<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Billy Mulley on Medium]]></title>
        <description><![CDATA[Stories by Billy Mulley on Medium]]></description>
        <link>https://medium.com/@billymulley11?source=rss-956d94bd874b------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*Y5jWXasPCE5wUo3d3ENlNQ.jpeg</url>
            <title>Stories by Billy Mulley on Medium</title>
            <link>https://medium.com/@billymulley11?source=rss-956d94bd874b------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Fri, 20 Nov 2026 00:29:06 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@billymulley11/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[Heidenheim’s 23/24 Set-Piece Prowess — Bundesliga Analysis]]></title>
            <link>https://medium.com/@billymulley11/heidenheims-23-24-set-piece-prowess-bundesliga-analysis-361d842e42be?source=rss-956d94bd874b------2</link>
            <guid isPermaLink="false">https://medium.com/p/361d842e42be</guid>
            <dc:creator><![CDATA[Billy Mulley]]></dc:creator>
            <pubDate>Thu, 04 Dec 2025 14:16:46 GMT</pubDate>
            <atom:updated>2025-12-04T14:16:46.075Z</atom:updated>
            <content:encoded><![CDATA[<h3><strong><em>Heidenheim’s 23/24 Set-Piece Prowess — Bundesliga Analysis</em></strong></h3><p><strong><em>“I was seen as a dinosaur for my focus on dead-ball situations and long-throws with Stoke City when we were promoted to the Premier League in 2008, but I wouldn’t say I feel vindicated by the way they are now in fashion — because I knew back then how important they were.” — Tony Pulis </em></strong><a href="https://www.bbc.co.uk/sport/football/articles/cx2p90x89pwo"><strong><em>(BBC Sport)</em></strong></a></p><p>Football moves in cycles. Set-pieces are sexy once again and you can certainly feel their renewed sense of importance in the modern game. Set-piece-specific coaching and analysis positions are growing increasingly popular, and data has the potential to transform how we see dead-ball situations. If you need a further indication of where set-piece analysis may be heading, then the fact that Sports Data Campus are starting an <a href="https://english-programs.sportsdatacampus.com/msc-set-pieces-in-football/">MSc Set-Pieces in Football Degree </a>might help.</p><p>Set-pieces represent a huge opportunity for analysts to make an impact. They start from a fixed and repeatable ‘dead-ball situation’, so there is subsequently a higher level of control over how they are designed, implemented, monitored and evaluated.</p><p>Attempting to highlight the importance of set-pieces and to discover patterns to explore in more detail, I have accessed raw event data from the 23/24 Bundesliga season (big thanks to <a href="https://github.com/ImpectAPI/open-data">Impect for releasing a fantastic open dataset</a>, which is a fantastic starting place for those building analysis projects and wanting to understand how different providers structure their datasets).</p><p>Into the analysis and 30.1% of the total goals during the 2023/24 Bundesliga came courtesy of a set-piece, which is a significant chunk considering that football is a low-scoring sport where fine margins often determine the outcome of matches. Similarly, shots from set-pieces and xG from set-pieces accounted for 30.6% and 29.3% of total shots and xG.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*tfQWw6hyRqHMYqXz-tiEEg.png" /><figcaption>Figure 1 — Set-Piece Goals vs Set-Piece xG</figcaption></figure><p>Figure 1 compares set-piece goals with set-piece xG within the Bundesliga 23/24 teams and indicates which teams were over- and underperforming their expected goals. Bayern Munich and Bayer Leverkusen amassed 25 and 14 goals from set-pieces respectively, with both significantly overperforming their xG. Interestingly, Mainz 05 accumulated the highest xG in the division (21.75), but managed just 16 set-piece goals.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*TzExvQObuL2rCtgU3g0L9Q.png" /><figcaption>Figure 2 — Set-Piece Contribution Across Goals, Shots, and xG</figcaption></figure><p>Figure 2 continues to highlight the power of set-piece situations on a team level. Notably, 53.6% of all FC Koln’s goals came from set pieces, although they scored a mere 28 goals in total. For VFL Bochum, Union Berlin and FSV Mainz 05, set-piece goals accounted for over 40% of total goals. Conversely, a mere 14.6% of all Eintracht Frankfurt’s goals came from set-pieces, despite their set-piece situations accounting for 29.7% and 27.7% of their total shots and xG.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*cK3UG_w6fBy8c9ACf56Rdw.png" /><figcaption>Figure 3 — Set-Piece Type Split for xG, Shots and Goals</figcaption></figure><p>Going a level further in examining how chances and goals were created from set-piece situations, Figure 3 outlines the set-piece type responsible for generating xG, Shots and Goals. The eye is immediately drawn to corners being responsible for around half of the split across the three charts, which is what was probably to be expected. Personally, the percentages for throw-ins across the three were surprisingly high.</p><h4>Heidenheim’s success fueled by set-pieces</h4><p>One narrative that is not abundantly clear but is certainly worth chasing is Heidenheim and their higher level of dependence on set-pieces for generating chances. FCH’s 8th-placed finish and subsequent Conference League qualification in what was the club’s first-ever season in the German top-flight was slightly overshadowed by Bayer Leverkusen’s first-ever title win, where they stormed the league by 17 points and went the season unbeaten.</p><p>Heidenheim’s set-piece prowess during the early part of the 23/24 campaign was highlighted in an article by <a href="https://totalfootballanalysis.com/set-piece-analysis/heidenheim-202324-set-piece-tactical-analysis-tactics">Marcel Seifeddine (Total Football Analysis)</a>. FCH continued to thrive and threaten from dead-ball situations and secured a Conference League spot via goal difference. At the start of the season, survival would have represented success, and by the end of it, they were preparing for Europe.</p><p>Going back to Figure 2, Heidenheim accumulated the highest % for set-piece generated shots and third for set-piece-generated xG. There is a slight drop-off (7th in the division) regarding set-piece-generated goals.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*WZY_MMElJh6dUSbrqeM0JQ.png" /><figcaption>Figure 4 — Heidenheim Set-piece type split</figcaption></figure><p>Naturally, with corners making up a significant share of their set-piece xG, shots and goals, it is the perfect place to start. Heidenheim amassed the fifth-highest number of corners during the 23/24 Bundesliga campaign, and Figure 5 illustrates the zones targeted from corners. Interestingly, Jan-Niklas-Beste’s wand of a left foot was tasked with delivering 78% of FCH’s corners, with the total of corners delivered by left-footers rising to 82%. As evidenced below, corners from the left corner (which will predominantly have been outswingers) reached deeper zones and inswingers from the right tended to target the near post.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*SkbjvGDlPTPs_CgpaoM-yg.png" /><figcaption>Figure 5 — Corner end locations for FC Heidenheim</figcaption></figure><p>Figure 6 goes on to visualise ‘successful corners’ and unsurprisingly, short corners take up a higher share, given that a teammate is much more likely to make first contact compared to a delivery into the box. What we do see, however, is that a third of successful outswinging corners are met centrally between the 6 and 18-yard boxes. From the right side, which are mostly inswingers, it is in fact the back post zone within the 6-yard box that is most productive. That is despite Figure 5 indicating that the near-post and central zones (within the 6-yard box) were targeted 54.6% of the time.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*5IJ-JXxfFkQwjFDyxW2sjQ.png" /><figcaption>Figure 6— Corner end locations for FC Heidenheim (successful corners only)</figcaption></figure><p>Going one step further in Figure 7, we factor in the quality of chances created and goals that have been scored from corners in each zone. Now, coaches and set-piece analysts can start working on next opposition analysis to implement strategies to protect highly targeted and previously fruitful areas. Interestingly, all four of Heidenheim’s goals from right-sided corners were scored inside the box and inside that central zone. Additionally, league-wide or even data across several divisions will help to identify trends and could aid Heidenheim in threatening further from corners.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*mIE1Bevl4Zn8i2g37wL5fQ.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*LjHa0CTuId219JM7aZbWEQ.png" /><figcaption>Figure 7 — xG accumulated and Goals Scored across pitch zones</figcaption></figure><p>This certainly has the potential to assist analysis workflows, but we could garner even more value if we start to look at things on an individual level. During the 23/24 season, Heidenheim’s Tim Kleindienst (now at Borussia Monchengladbach) accumulated the third-highest xG from corners across the division, as evidenced in Figure 8. He also took 13 shots during corner sequences, with teammate Patrick Mainka managing one more.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*AV3VzLTllrTEwPUSxPKP4A.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Ui4zgNE2jgt_g4lc8plmEw.png" /><figcaption>Figure 8 — xG and Shots from corners at player level (Bundesliga 23/24)</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*hIRHwhslKK3QNaJg8xCXfQ.png" /><figcaption>Figure 9— Tim Kleindienst’s Shot and Goal locations from corners (Bundesliga 23/24)</figcaption></figure><p>On the flip side, Kleindienst proved to be just as, or even more impactful in defending corners. He managed the most defensive actions from corners from the opposition (51), with 43 of those being in the six most central zones.</p><h4>Conclusion/next potential steps</h4><p>Starting with raw event data from Impect, and subsequently building pipelines and metrics that are tailored towards analysing set-pieces, this article only scratches the surface of what can be done.</p><p>This project honed in on one particular team and one particular set-piece type, meaning there is a tremendous amount of data that has not been explored.</p><p>I also believe deeper analysis on set-piece sequences could be a good next step. Are short corners more valuable? Can we visualise throw-in sequences to identify patterns?</p><p>Given the wealth of information available and the multitude of insights that can be generated, I’d be interested in doing something similar, but with a brief from a coach. It is quite easy to get carried away in analysing aspects that may not be too rewarding, whilst speaking to coaches would probably open the door to more tailored, unique analysis that is more likely to benefit a club.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=361d842e42be" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Generating insights from Copa Libertadores 2025 shot data]]></title>
            <link>https://medium.com/@billymulley11/generating-insights-from-copa-libertadores-2025-shot-data-f1dc91ee5d2f?source=rss-956d94bd874b------2</link>
            <guid isPermaLink="false">https://medium.com/p/f1dc91ee5d2f</guid>
            <dc:creator><![CDATA[Billy Mulley]]></dc:creator>
            <pubDate>Wed, 08 Oct 2025 14:06:14 GMT</pubDate>
            <atom:updated>2025-10-08T14:18:20.494Z</atom:updated>
            <content:encoded><![CDATA[<p>Free, structured and high-quality datasets are hard to come by in the world of football data, which is why when <a href="https://www.linkedin.com/in/axel-alvarado/">Axel Alvarado</a> posted comprehensive shot data from the 2025 Copa Libertadores group stages, it felt like a perfect opportunity to explore further.</p><p>With each shot being accompanied by a wealth of information like the location through x and y coordinates, whether the effort was blocked or not, xG, goalkeeper information, etc, there are so many ways to generate insights from this dataset!</p><p>First of all, Figure 1 takes a look at the players who have outperformed their xG.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*ggrG4uFa9MrxlJxxfXp-CQ.png" /><figcaption>Figure 1 — Goals vs xG dumbbell chart — Top 10 players</figcaption></figure><p>Being a competition with only a small number of games, it is no surprise to see such distances between the top scorers’ goals and their xG. Even so, it is hard to ignore Bolivar’s Ramiro Vaca’s incredible strike rate, managing 5 goals from a mere 0.49 xG.</p><p>Honing in on the 26-year-old, I created a shot map to better visualise Vaca’s shooting locations, and it is safe to say, he loved a shot from distance during the group stages of Copa Libertadores 2025. Interestingly, 14 of his 18 attempts at goal came from outside of the area, and all 18 were with his stronger right.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/702/1*fMVWl4wL7jrkX9PZQ1lEtA.png" /><figcaption>Figure 2 — Ramiro Vaca’s Copa Liberadores 2025 shot map</figcaption></figure><p>Set-pieces appear to be gaining renewed emphasis in modern football, reflected in the number of clubs hiring dedicated set-piece coaches and analysts.</p><p>In line with there being a heightened level of focus on these set-piece situations, Figure 3 presents the top 10 players who have generated the most xG from attacking corners.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*11xCanXTexlmxlkAEX0uRw.png" /><figcaption>Figure 3 — Top 10 Players by xG from Att corners</figcaption></figure><p>It is two Sebastians (Driussi and Coates) who lead the way here, with both players accumulating over 1xG from attacking corners.</p><p>This graphic, or a slight variation that honed in on one particular team, would be beneficial to opposition scouts, who could help shape a strategy to combat these danger men from continuing to threaten from attacking corners.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*RN-Ag4FU7-hQFmvNH5IhCw.png" /><figcaption>Figure 4 — Top 10 Teams by xG from Set-Pieces</figcaption></figure><p>Figure 4 illustrates which teams generated the most xG from set pieces. Interestingly, across six group games, River Plate accumulated nearly 7 xG from set pieces — accounting for over 50% of their total xG. As shown in the visual, Libertad and Universidad de Chile also recorded higher xG from set pieces than from open play.</p><p>Thank you once again to Axel Alvarado for building and providing this dataset, with this article only scratching the surface of what can be analysed!</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=f1dc91ee5d2f" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Junnosuke Suzuki — Using raw event data to analyse F.C. København’s recent addition]]></title>
            <link>https://medium.com/@billymulley11/junnosuke-suzuki-using-raw-event-data-to-analyse-f-c-k%C3%B8benhavns-recent-addition-4ed833ce433c?source=rss-956d94bd874b------2</link>
            <guid isPermaLink="false">https://medium.com/p/4ed833ce433c</guid>
            <dc:creator><![CDATA[Billy Mulley]]></dc:creator>
            <pubDate>Fri, 01 Aug 2025 16:37:46 GMT</pubDate>
            <atom:updated>2025-08-01T16:37:46.131Z</atom:updated>
            <content:encoded><![CDATA[<h3>Junnosuke Suzuki — Using raw event data to analyse F.C. København’s recent addition</h3><p>My previous projects have begun using standard, pre-designed metrics from established and respected data providers. While these providers play a vital role in football’s data analytics landscape, I wanted my next project to challenge me in a different way.</p><p>This article, therefore, explores working with raw event data, transforming it into commonly used metrics (and a couple of more advanced metrics), and then attempting to extract meaningful insights from the ground up.</p><p>This project uses<a href="https://www.hudl.com/blog/j1-league-free-data-statsbomb"> Hudl StatsBomb’s free 2024 J-League dataset</a>. Statsbomb have succeeded in ‘supporting the development of the next generation of analysts”, and their continued provision of free data has certainly helped me learn, experiment, and improve. Interestingly, it includes advanced physical metrics, but integrating those is something I’m keen to explore in a future project.</p><p>A few months ago, when I first accessed this dataset, I worked on a project that I was proud of at the time, as it was my first experience producing something meaningful with raw event data, but it was something I knew I wanted to revisit once I felt I could improve the work.</p><h4>The Process</h4><p>Being such a large and complex dataset, I spent a long time exploring and experimenting with the data.</p><p>Using event data like substitutions, red cards, and half-time timestamps, I built a function to calculate how many minutes each player spent on the pitch in each match, which enabled me to generate per-90 metrics further down the line. I also cleaned and structured location (x,y) data in a way for easier access.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*5b8KiEvQjLwn5QaritHUBQ.png" /><figcaption>Figure 1 — Example of how two metrics were constructed using Python</figcaption></figure><p>Next, I created fairly standard metrics, with two of them being pictured above. All metrics were defined before merging them into a master dataset. The final two steps of this part of the process were normalising the appropriate metrics (per 90), filtering for players who had amassed over 500 minutes, and filtering for different positions.</p><h4>Analysis</h4><p>At this point, I still did not have an angle or a plan of where I wanted the project to go.</p><p>Then, F.C. København announced the arrival of highly-rated defender Junnosuke Suzuki.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*sikrs1z-CEvMHtZHHqywFA.png" /><figcaption>Figure 2 — Junnosuke Suzuki’s 2024 J-League Percentile Rank Performance compared to CBs</figcaption></figure><p>Of course, the immediate drawback in all of this is that this project only captures and focuses on the 2024 J-League season. Suzuki continued to rise to prominence during the 2025 campaign, and he even made his senior international debut for Japan in June. However, this project still manages to highlight what an impressive talent Suzuki is. As evidenced by Figure 2, Suzuki ranked very well in several metrics, particularly from a ball progression standpoint.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/990/1*IfxoU-rNAQh99Q-fZcLdTQ.png" /><figcaption>Figure 3 — ‘Hybrid CB’ Z-Score</figcaption></figure><p>Running a Player Rating model to assess how Suzuki performed as a ‘Hybrid CB’ (<a href="https://medium.com/@billymulley11/chatgpt-give-me-a-football-scouting-mission-using-data-to-create-a-first-step-in-the-e825cdc6785e">same methodology as this project here)</a>, he actually comes out on top against his positional counterparts. To generate a ‘Hybrid CB’ profile, I applied composite scoring with relatively neutral weightings across selected metrics.</p><h4>Defensive analysis</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*KK0WeZlpkzGxy77PDQq2mg.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*TqnXGtuWm5Ut4RvpwfpOCQ.png" /><figcaption>Figure 4 — Junnosuke Suzuki’s Tackle Success (%) and Aerial Duel Win (%) by zone</figcaption></figure><p>As Figure 2 shows, Suzuki ranked within the top 3% for Tackles Won per 90, but the spread of where these tackles were won makes for interesting viewing, especially when compared to the locations of Aerial wins (Figure 4). A sizeable number of the young defender’s tackles were actually high up the pitch.</p><p>Aerially speaking, he tends to compete in much deeper areas of the pitch. Interestingly, he has competed aerially in the wide left channel much more than I would have anticipated, seeing a varying amount of success in doing so.</p><p>As the pizza chart suggested, Suzuki underperformed in the number of aerial duels that he won and his success rate. Taking this one step further and starting to analyse his ability to defend the box in a bit more detail, I focused on the deepest central zone.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/990/1*ozFImFXwVEpwog9nL5xRaQ.png" /><figcaption>Figure 5 — Aerial Success Defending Central Box Zone</figcaption></figure><p>As per Figure 5, Suzuki ranked very lowly across these two metrics in this particular zone. This is something I’d be keen to compare against his 2025 performance, to better determine if this is an aspect of his game that has improved.</p><p>Interestingly, he ranked 70/77 for his Aerial Win success rate (%) from all J-League CBs who had accumulated over 500 minutes. He also averaged a mere 0.78 Aerial Wins per 90 in this zone.</p><h4>Progression analysis</h4><p>Ranking in the top 6% for Progressive Carries and top 7% for Progressive Passes, his ability to advance the ball is clearly a standout trait.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*PX8L9In5ViDLnxIcglBdzw.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*iN1qzqmOAKWhH-D_uMrlMQ.png" /><figcaption>Figure 6— Locations of Progressive Passes and Carries by zone (accompanied by count of where action originates)</figcaption></figure><p>Unsurprisingly, a significant share of these progressive actions originate from the left half-space. For passes specifically, his most effective zone was just behind the halfway line. However, Figure 6 displays that he still managed to deliver an impressive 19 progressive passes from a more challenging zone just ahead of the halfway line, where forward options are more restricted.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*9QCqMUacTD1IMN2VLMaU9A.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*X9_ZS4Rt7jxDipz9BTGVSA.png" /><figcaption>Figure 7 — Junnosuke Suzuki’s Pass + Carry OBV by Zone</figcaption></figure><p>Figure 7 is where we can start to hone in on where on the pitch Suzuki adds the most value with his progressive actions. We can calculate how much On-Ball Value (OBV) the young defender positively or negatively contributes through his carries and passes. From a Pass OBV standpoint, it is out wide on the left, and in slightly more advanced areas, where he particularly impresses. It is from deeper positions in the left half spaces where Suzuki adds more carrying value.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*wW4Hy25P92sMLVafdb-Nsg.png" /><figcaption>Figure 8 — Junnosuke Suzuki’s Carry OBV per 90 vs average J-League (2024) LCB</figcaption></figure><p>Attempting to go one step further and being particularly intrigued by his carrying ability, figure 8 provides further insight into the value Suzuki has added through his carries. Normalising Carry OBV by calculating it as a per 90 metric and filtering specifically for left-sided centre-backs, we can start to compare how the Japan international compared to his positional counterparts. The percentage represents the percentage difference between Suzuki’s zonal Carry OBV against the LCB average.</p><p>Figure 8 provides some interesting insights. Firstly, the left half space continued to be a productive zone for Suzuki, particularly in more advanced positions outside of the penalty area. The central zones are also strong zones for Suzuki.</p><h4>Conclusion</h4><p>Starting from raw event data, this project has enabled me to be more creative and experimental with the insights produced. Splitting the pitch into zones has helped provide further context regarding where on the pitch he is highly involved, which I believe has been particularly interesting when honing in on Suzuki’s progressive actions.</p><p>As alluded to, an obvious extension to this would be to access data from the current J-League campaign. This, in combination with the 2024 season data, would allow us to track his progress further and outline how his game has changed across the two seasons.</p><p>Suzuki’s move to F.C. København is an exciting one and I look forward to seeing how he adapts to his new surroundings in what is an incredibly exciting opportunity to continue his development.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=4ed833ce433c" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Clustering Premier League Midfield Profiles — Exploring a Different Data-Driven Method to Initiate…]]></title>
            <link>https://medium.com/@billymulley11/clustering-premier-league-midfield-profiles-exploring-a-different-data-driven-method-to-initiate-cee99bf5f0f2?source=rss-956d94bd874b------2</link>
            <guid isPermaLink="false">https://medium.com/p/cee99bf5f0f2</guid>
            <dc:creator><![CDATA[Billy Mulley]]></dc:creator>
            <pubDate>Wed, 16 Jul 2025 16:24:15 GMT</pubDate>
            <atom:updated>2025-07-16T16:24:15.449Z</atom:updated>
            <content:encoded><![CDATA[<h3>Clustering Premier League Midfield Profiles — Exploring a Different Data-Driven Method to Initiate the Recruitment Process</h3><p>Data is driving innovation in modern football, as more clubs embrace data-led strategies. Machine learning is becoming particularly impactful in the modern game, influencing several functions within a club. From performance analysis and recruitment to marketing and fan engagement strategies, data science is growing increasingly prominent.</p><p>As a scout, who is intrigued by football’s relationship with data and how advanced analytics can shape Talent Identification, I wanted my next project to explore how machine learning methods can be integrated into the recruitment process.</p><p>Building on my <a href="https://medium.com/@billymulley11/chatgpt-give-me-a-football-scouting-mission-using-data-to-create-a-first-step-in-the-e825cdc6785e">first Medium article</a>, this piece explores another approach to initiating the scouting process, specifically, how clustering can help narrow a long-list into a shorter, more manageable short-list.</p><p><strong><em>Research question: </em></strong><em>Can a clustering method be applied to football recruitment to narrow the pool of players being assessed?</em></p><p>There is a lot of variation when it comes to the individual role types that exist across positions. Rodri and Kevin De Bruyne may be two world-class midfielders, but the pair offer a completely different service and possess unique attributes that enable them to excel within their respective roles.</p><p>The idea behind clustering, in this particular case, was to identify players who operate with similar responsibilities and display similar attributes, in order to avoid comparing players with differing functions. This is in line with the author of <a href="https://www.amazon.co.uk/How-Win-Premier-League-Revolution/dp/152993463X">‘How to Win the Premier League’</a> and former Director of Research at Liverpool, Ian Graham’s pursuit of comparing “apples with apples” within his Possession Value model.</p><h4>The Process</h4><p>Using Premier League 2024/25 event-level data from Wyscout, I filtered for players who primarily operated as Midfielders or Attacking Midfielders, before selecting 30 different metrics to compare. Proceeding to standardise this data to aid further analysis, I converted my numerical data into z-scores.</p><p>From this point, I conducted Principal Component Analysis to transform these 30 metrics into more manageable and a smaller number of components, which still manage to capture a lot of the data’s variance. The objective of doing this was to later visualise clusters in 2 dimensions.</p><p>There are different ways to proceed with K-Means clustering, but knowing I wanted to separate players into 5 specific midfielder profiles, I was able to specify the number of clusters I wanted to create.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*lRojzR0EcXkxsoImLUKcPg.png" /><figcaption>Figure 1 — Heatmap of cluster profiles by metric</figcaption></figure><p>Figure 1 provides insight into the individual components of each cluster. Inspired by the <a href="https://youtu.be/ZtY8PnWY1Ek?si=5tYDCtups_cBD9o1">‘Clustering Women’s Football Players’</a> at the 2023 Statsbomb Conference, this visualisation allows us to start labelling the profile of each cluster.</p><p><strong><em>Cluster 0: Ball-Winning Destroyer<br> Cluster 1: Box-to-Box Midfielder<br> Cluster 2: Deep-Lying Playmaker<br> Cluster 3: Advanced Chance Creator<br> Cluster 4: Advanced Goal Threat</em></strong></p><p>Having identified distinct midfield role types through clustering, we can now visualise them in 2 dimensions.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*nHOFhVPd7FQRRsXO1Teabw.png" /><figcaption>Figure 2 — Premier League Midfielders and Attacking Midfielders split into clusters in 2 dimensions</figcaption></figure><p>Figure 2 visualises the different midfielder role types that exist within the Premier League. I’ve added labels to four players per cluster to better determine if the visualisation meets what we expect to see.</p><p>Of course, this is where an element of subjectivity kicks in, but there is nothing outstanding here that suggests this model is outright poor. Martin Odegaard’s proximity to Kevin De Bruyne and Mikkel Damsgaard follows expectations. Alexis MacAllister is classified as a Deep-Lying Playmaker, Morgan Rogers as an Advanced Goal Threat, so I am confident in the model’s performance from a quick glance.</p><p>However, there are a few things to consider at this point. Firstly, my calculation shows that the first two principal components explain 46.4% of the total variance, meaning the 2D visualisation captures less than half of the dataset’s variability. So, there will be some patterns or distinctions between midfield roles that may not be fully visible.</p><p>Secondly, the rather sharp distinctions between players and their midfield role present an immediate limitation with our objective in mind.</p><p>Let’s use Morgan Gibbs-White as an example, who has been classified as a Box-to-Box Midfielder. Although he is assigned to the Box-to-Box Midfielder cluster, Gibbs-White lies in close proximity to Phil Foden and the centroid of the Advanced Chance Creator cluster.</p><p>Considering the role he plays and the service he provides at Nottingham Forest, the eye test would suggest that he is more of an Advanced Chance Creator. MGW almost exclusively operated as an AM in a 4–2–3–1 formation during the 24/25 season, and perhaps Forest mastering the art of defending in a low-block has resulted in the 25-year-old contributing more defensively than your typical ACC.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*UFfL10aOv749adBWaU52bQ.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*WRr6L2jZYNoPr4ANfYxDmQ.png" /><figcaption>Figure 3 — Morgan Gibbs-White’s Percentile Rank performance amongst ‘Box-to-Box Midfielders’ vs ‘Advanced Chance Creators’</figcaption></figure><p>This highlights why we would need to be careful if a K-Means clustering model is used to filter players strictly by role. Figure 3 demonstrates exactly how different MGW performs against his counterparts in the two clusters (important to note that the sample size for Advanced Chance Creators was small — 9 players).</p><p>As a potential extension, a Gaussian Mixture Model could be used to provide probabilities of which cluster a player should belong to.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/780/1*YUb6Jb_k7WmEYxdcCONgfw.png" /><figcaption>Figure 4— Probability of Morgan Gibbs-White featuring in each cluster</figcaption></figure><p>At this point, we can start to devise a strategy that combats the aforementioned limitation of there being strict divisions between role types. Perhaps we can expand our player pool by including players whose percentage of appearing in the Advanced Chance Creator cluster is above 20%.</p><h4>Conclusion</h4><p>I believe this clustering method has the potential to be a starting point in the recruitment process. Of course, it is important to consider a clustering model’s drawbacks, but there is scope to be flexible and adapt the methodology to combat these limitations.</p><p>These types of articles are a great opportunity for me to develop my ability to apply data science in a football context. So, I’d really value any thoughts, suggestions or general feedback.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=cee99bf5f0f2" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Does stylistic variation still exist in modern English football?]]></title>
            <link>https://medium.com/@billymulley11/does-stylistic-variation-still-exist-in-modern-english-football-23639ad209e2?source=rss-956d94bd874b------2</link>
            <guid isPermaLink="false">https://medium.com/p/23639ad209e2</guid>
            <dc:creator><![CDATA[Billy Mulley]]></dc:creator>
            <pubDate>Mon, 07 Jul 2025 14:26:26 GMT</pubDate>
            <atom:updated>2025-07-08T09:13:03.687Z</atom:updated>
            <content:encoded><![CDATA[<p>As I’ve started writing more regularly, I&#39;ve found myself drawn to a recurring, broader thought about one of the most transformative figures in modern English football: Pep Guardiola. His influence is unmistakable and appears to extend far beyond the Premier League. Across the English football pyramid, we increasingly see teams adopting a more ball-oriented, possession-based attacking approach — each attempting, in their own way, to emulate a style that has proven highly effective at the elite level. This piece explores how deeply that influence runs.</p><p>Looking solely at his achievements at Manchester City, he has elevated the concept of dominance to an entirely new level. Meticulous from a tactical standpoint, Pep has developed players into exceptionally intelligent decision-makers on and off the ball — a key factor behind City’s six Premier League titles in the last eight seasons.</p><p>However, I wanted to investigate the current English footballing landscape to better understand the extent to which Guardiola’s more measured, strategic style of play has trickled down the pyramid. Is there a less vertical, more patient game in League 2 that revolves around spatial control? What is the relationship between attacking approach and chance generation? Does stylistic variation still exist in the modern era?</p><p>Setting out to explore this further, I focused on three freely available metrics <a href="https://theanalyst.com/competition/premier-league/stats">(provided by Opta Analyst)</a>: Direct Speed, the number of 10+ Pass sequences, and Open Play Expected Goals. My analysis covered the following leagues: Premier League, Championship, League One, League Two, Bundesliga, La Liga, and Ligue 1 (all from the 24/25 season).</p><p>Of course, honing in on last season’s data does not provide me with an accurate reflection of how the game has been played across different seasons, nor do the metrics selected fully capture the stylistic complexities that are present in today’s game. What we do have, however, is something of a foundation to begin comparing broader and current tactical differences across leagues and levels.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/985/1*HJRoHB5ITfJTmDv2HKAGBg.png" /><figcaption>Figure 1 — Direct Speed vs Open Play xG (Premier League 24/25)</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/984/1*pRY2-2XcmMVfUsWhm-cOXg.png" /><figcaption>Figure 2— 10+ Passes vs Open Play xG (Premier League 24/25)</figcaption></figure><p>Figures 1 and 2 illustrate the correlations between the two metrics and chance creation in the Premier League, as measured by the Pearson Correlation Coefficient. The correlation between Direct Speed and Open Play xG is extremely low, and perhaps more surprising is the fact that it is positive. We see a stronger relationship between the count of 10+ Passes and xG in Figure 2, equating to an R-squared value of 0.26, suggesting 26% of the variance in Open Play xG can be explained by the count of 10+ Passes.</p><p>One particular talking point is the distance between Liverpool and Manchester City across both figures, especially when considering the narrative around Arne Slot bringing a more measured, patient style of football to Anfield. Bournemouth’s comparatively low pass-sequence and faster style of football is equally as intriguing, considering they generated the fifth-highest Open Play xG.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/984/1*AHqId5ub7-TPaldCu4VikQ.png" /><figcaption>Figure 3— Direct Speed vs Open Play xG (League 2 24/25)</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/986/1*nfoyceziEmIoiMLO0XHoaQ.png" /><figcaption>Figure 4– 10+ Passes vs Open Play xG (League 2 24/25)</figcaption></figure><p>What we see when we look at the two metrics across League 2 is very interesting. Across both figures 3 and 4, it is Notts County, Chesterfield and MK Dons, who stand out considerably, with the three teams adopting a ball-oriented, patient style of attacking. Something that is equally intriguing is that those three teams have enjoyed a varying level of chance-creation success.</p><p>It is important to once again bring up that these two metrics alone do not cover the tactical nuances that exist in the division. However, the spread of the data in Figure 3 is indicative of the stylistic variation that remains present in England’s fourth tier.</p><p>I am a fan of tactical variation. I like that there are still plenty of success stories from teams who stylistically approach the game differently. I appreciate that longer, more direct football remains prevalent and effective in certain contexts. At the same time, I also find myself in agreement with <a href="https://www.tntsports.co.uk/football/champions-league/2022-2023/pep-guardiola-has-transformed-english-football-from-backwards-to-technical-say-bt-sport-pundits_sto9211601/story.shtml">comments from TNT Sports pundits </a>Rio Ferdinand and Glenn Hoddle, who have indicated that Pep Guardiola’s influence on the English game has trickled down to how academy football is coached, empowering players in this country to have more touches of the ball and subsequently reach an elevated technical level.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/984/1*RbPYxVzklClAJQbY_5fcWw.png" /><figcaption>Figure 5— Direct Speed vs Open Play xG (Bundesliga 24/25)</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/986/1*LjlRiVq9U7tyXhFnuY6ASw.png" /><figcaption>Figure 6–10+ Passes vs Open Play xG (Bundesliga 24/25)</figcaption></figure><p>While the primary focus of this research was to assess the impact of the aforementioned metrics within the context of English football, it is also valuable to explore and visualise these trends in a different setting. The strongest correlations between the metrics and xG were observed in the Bundesliga, making it an ideal division to examine more closely.</p><p>Figures 5 and 6 suggest Bayern were in a league of their own regarding attacking approach. Eintracht Frankfurt, VfB Stuttgart, Borussia Dortmund, and Bayer Leverkusen then appear in a secondary bracket, particularly noticeable in Figure 6.</p><p>There is a strong statistical relationship between the count of 10+ Passes and xG, as shown in Figure 6. 74% of the variance in xG can be explained by the number of longer passing sequences. This suggests that possession-based buildup is closely linked to chance creation, but this is still a statistical association and does not imply causation.</p><h4>Possible Extensions:</h4><p>As alluded to, I was only able to access data from the 24/25 season, preventing me from assessing how the landscape has changed over time and identifying any meaningful trends. The two selected metrics do play a large role in understanding</p><p>As an extension to this, I would be interested in visualising how market value/budget impacts how teams in different divisions construct their attacks. I believe the growing financial disparities that exist within divisions can somewhat dictate style, so that would be a potential avenue to explore further.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*eYlfSYgP3GezcGLjQ_xNmA.png" /><figcaption>Figure 7 — Premier League, Championship, League 1, League 2, La Liga, Bundesliga, Ligue 1 ranked by Direct Speed (24/25)</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*bYmy1v7V4MZRZYh18Z-yyw.png" /><figcaption>Figure 8— Premier League, Championship, League 1, League 2, La Liga, Bundesliga, Ligue 1 ranked by count of 10+ Passes (24/25)</figcaption></figure><p>Figures 7 and 8 highlight the 10 teams with the slowest Direct Speed and the 10 teams with the highest number of 10+ pass sequences within the dataset. While there may be a few surprise inclusions, the majority of teams align with expectations.</p><p>Exploring data from European competitions would have also added valuable context. PSG ranked first for sequences of 10+ Passes and 5th in Direct Speed within their respective division, which is completely understandable when considering the financial gulf between themselves and their divisional counterparts. Had we been able to explore these metrics in the context of the Champions League, we could have assessed whether their measured, ball-oriented style persisted in a setting where they did not hold the same financial advantage.</p><p>Of course, PSG went on to win the competition, exhibiting an innovative and fluid style of football that perhaps reassured fans of the beautiful game that the game is still beautiful. However, I think we saw tactical variation across their Champions League campaign, proving to be slightly more adaptable in certain settings. For a deeper tactical breakdown of PSG’s recent evolution, see <a href="https://olioconnell.substack.com/p/the-new-age-of-football?utm_campaign=post&amp;utm_medium=web">Oli O’Connell’s article</a> on Les Parisiens’ European triumph, which examines how the club has embraced social dynamics and tactical innovation over reliance on individual talent, all whilst bringing the concept of diagonality to prominence.</p><p>Direct Speed and long passing sequences are useful starting points, but they only scratch the surface of what defines a team’s style of play. There is certainly scope for more advanced, tailored metrics to be generated, which can capture stylistic complexity in greater detail. At club level, analysts will adapt, bring across or create metrics that will enable them to analyse how well their team adheres to a pre-defined style of football. So, it would be interesting to measure performance against tactical objectives and monitor the impact of external factors on strategy over the course of a season.</p><p>Pep has been an undeniably positive influence on English football. After the 2024/25 season — which he will surely hope is just a temporary setback — I’m intrigued to see the strategy behind the response for the upcoming Premier League campaign.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=23639ad209e2" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Identifying Potential in the ‘League With the Most Undervalued Talent’]]></title>
            <link>https://medium.com/@billymulley11/identifying-potential-in-the-league-with-the-most-undervalued-talent-e253ec7c5f2f?source=rss-956d94bd874b------2</link>
            <guid isPermaLink="false">https://medium.com/p/e253ec7c5f2f</guid>
            <dc:creator><![CDATA[Billy Mulley]]></dc:creator>
            <pubDate>Tue, 01 Jul 2025 08:58:58 GMT</pubDate>
            <atom:updated>2025-07-01T08:58:58.248Z</atom:updated>
            <content:encoded><![CDATA[<p>Data analytics is playing an increasingly important role in the recruitment process, shaping how talent is identified and evaluated. Complex recruitment algorithms now stretch way beyond the ability to locate talent based on goal output.</p><p>Now, advanced machine learning models can input millions of data points to produce tailored, wide-ranging and instantaneous insights that provide decision-makers with a more complete overview of a player, their current and potential values and stylistic fit within different tactical set ups.</p><p>For British clubs, eased restrictions and increased flexibility within the FA’s Government Body Endorsement (GBE) regulations have enabled teams to access markets that were previously difficult to navigate, and it has subsequently improved clubs’ scouting coverage beyond the traditional leagues where eyes were previously fixed.</p><p>Of course, there are still regulations regarding how many GBE slots teams at different levels are allocated, but it has certainly encouraged clubs to be more creative in how they approach their recruitment strategy.</p><h4>Risk minimisation by searching in undervalued markets</h4><p>Naturally, exploring untested or underexplored markets represents a significant risk. That is because there are limited case studies for decision-makers to analyse and perhaps integrate into a model.</p><p>However, sticking to the same talent pools inevitably increases demand, and as demand grows, prices start to inflate, making it harder to find true value.</p><p>Looking at the darlings of data analytics in English football, Brighton and Brentford have reaped the rewards of a brave, data-backed recruitment strategy. Being early adopters of advanced analytics has been transformative and has accelerated the development, stature and global reputation of both clubs. While there are notable differences in the markets they source talent from, the common denominator is clear: they consistently extract value where others have overlooked it.</p><p>With that in mind, this latest project hones in on talent within the Czech First League, following research from <a href="https://www.twentyfirstgroup.com/tfg-world-sports-rankings/quality-rankings/?sr=rank-football-2">Twenty First Group in December 2024</a>, which highlights the division as the most undervalued men’s competition.</p><h4>The Czech First League: Europe’s Underrated Talent Goldmine</h4><p>Twenty First Group’s research indicates that the Czech First League have been selling players at a 59% discount when considering a player’s ‘true worth’. Norway’s Eliteserien, Japan’s J-League, Sweden’s Allsvenskan and Super League Greece make up the rest of the top 5.</p><p>Going into the 2025/26 campaign, the Czech First League remains in band 4 , meaning it is quite difficult for players to attain a GBE autopass or amass 15 points in the criteria laid out by the FA. Still, there is plenty of scope for British clubs to go down the Elite Significant Contribution (ESC) route, which provides those players who do not accumulate the 15 required points with the ability to still make a move. There are a limited number of slots that clubs in different divisions are allowed to use. The best place to learn or keep up to date with the latest GBE/ESC regulations is the <a href="https://www.gbeexperthub.com/">GBE Expert Hub.</a></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/989/1*HR7KG17CFifGAA_saOQ5LQ.png" /><figcaption>Figure 1 — Competitions that have sourced Czech First League talent since 23/24 season</figcaption></figure><p>Figure 1 backs up the earlier comment that the Czech First League is an underexplored market. Since the 2023/24 season, the Austrian Bundesliga and the Danish Superliga lead the way in bringing in talent from the Czech top-flight. La Liga, Eredivisie, Bundesliga and 2. Bundesliga have taken two players each. Equally, the English Championship has also welcomed two players in Muhamed Tijani (Plymouth Argyle) and Matej Jurasek (Norwich City). Tottenham Hotspur are the only Premier League club to have recruited from the division, signing Slavia Prague goalkeeper Antonin Kinsky in January.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*GmZrQtYgjxfSgj9t_BGrTQ.png" /><figcaption>Figure 2 — Nationalities of departed Czech First League players since 23/24 season</figcaption></figure><p>Unsurprisingly, Figure 2 exhibits that almost two-thirds of all player departures have been Czechs themselves. Nigeria accounts for the second-highest share of Czech First League player departures (8.6%), with individuals from the likes of The Gambia, Equatorial Guinea and Cote D’Ivoire also featuring. This aligns with the views of MRKT Insights co-founder Tim Keech, who noted on the <a href="https://www.youtube.com/watch?v=v4FIxTP4sWU&amp;list=PLkukQbkffr_u0zaa_vvmx-5yerEfiw8w0&amp;index=28">Chris Gill Podcast </a>that Czech clubs have shown a strong appetite for investing in young African talent — an approach that has paid off both financially and in terms of on-field performance.</p><p>It will certainly be interesting to see how this trend develops. Still, there seems to be a level of hesitancy (for the most part) within the elite competitions when it comes to buying talent African talent, with teams in countries like Sweden, Norway, Denmark and of course the Czech Republic benefitting from essentially bridging the gap and generating huge profits in a short space of time.</p><p>That, in itself, is a topic for another day and something I am keen to revisit. For now, here are three intriguing prospects who could represent the next wave of Czech-based talent, potentially available at what many would consider a bargain price.</p><h4>Qëndrim Zyba</h4><blockquote>Club: <strong>Banik Ostrava</strong>| Age: <strong>24 </strong>| Profile: <strong>Box-to-Box Midfielder </strong>| Wyscout value: <strong>€500k</strong></blockquote><p>Up first is an incredibly well-rounded, all-action midfielder in 24-year-old Qëndrim Zyba. The Kosovo international accumulated just over 1,000 minutes of First League action during the 24/25 campaign, proving to be an integral part of his Slovan Liberec side when fit. After running his Wyscout data through my basic box-to-box midfielder algorithm — using z-score normalisation and composite scoring — he returns a leading ‘Compatibility Score’ across 67 midfielders who played more than 500 minutes.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/989/1*XHRWtkz2U_H5wlKPVDLwRQ.png" /><figcaption>Figure 3 — Qendrim Zyba’s Box-to-box midfielder Compatibility Score</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*P2r0FohckqEw8IDKSIeapQ.png" /><figcaption>Figure 4— Qendrim Zyba’s Percentile Rank Performance</figcaption></figure><h4>Tomas Rigo</h4><blockquote>Club: <strong>Banik Ostrava</strong>| Age: <strong>22 </strong>| Profile: <strong>Box-to-Box Midfielder </strong>| Wyscout Value: <strong>N/A</strong></blockquote><p>Tomas Rigo sticks out as another intriguing profile, who similarly returns impressive data using the same box-to-box midfielder algorithm. Amassing nearly 3,000 minutes of league action, he primarily operated as a CM in a 3–4–3 formation at Banik Ostrava, who achieved an impressive 3rd-placed finish during the 24/25 campaign.</p><p>Ranking 6/67 in regards to his Compatibility score, the 22-year-old played an integral role in his side’s positive season, registering a goal and six assists in the process.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/989/1*KHsaorXEVZoH47MTSfzRsg.png" /><figcaption>Figure 5 — Tomas Rigo’s Box-to-box midfielder Compatibility Score</figcaption></figure><p>Rigo returns very strong Progressive data, ranking in the 99th and 93rd percentiles for Progressive Runs per 90 and Key Passes per 90, respectively. Also, ranking in the 94th and 79th percentiles for xA per 90 and Assists per 90 showcase his creative abilities from a central position.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*PgO6AbKYRo5xm-oxhAXBGQ.png" /><figcaption>Figure 6 — Tomas Rigo’s Percentile Rank Performance</figcaption></figure><h4>Pavel Šulc</h4><blockquote>Club: <strong>Viktoria Plzen</strong> | Age: <strong>24</strong> | Profile: <strong>Inverted Winger</strong> | Wyscout Value: <strong>€6m</strong></blockquote><p>Turning attention to a player who has <a href="https://www.efotbal.cz/article/fotbal-chance-liga-sulcuv-klicovy-pulrok-sleduji-ho-tymy-z-top-lig-k-nabidce-se-ale-nemaji-zije-sance-na-senzaci-ab7209f3-189d-4dcb-8489-4afff64326f5">already generated transfer interest</a> from the likes of Manchester City, Tottenham Hotspur and Aston Villa, Viktoria Plzen attacker Pavel Sulc is a player who undoubtedly possesses an exciting future. Ran through an ‘Inverted’ winger algorithm, the 24-year-old played a similar number of minutes as a left winger as he did as an attacking midfielder.</p><p>A contingent of interested parties may spark somewhat of a bidding war and drive the price north of what would be considered good value, but there is still real potential for clubs to steal Sulc for a bargain fee.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/989/1*rO-0DpdHTiakdJfkfXVvJQ.png" /><figcaption>Figure 7 — Pavel Sulc’s ‘Inverted’ winger Compatibility Score</figcaption></figure><p>Honing in on Sulc’s Percentile Rank Performance, he returns strong and fairly well-rounded data. Excelling in his Key Passes per 90, Aerial metrics, PAdj Interceptions and Goal Output metrics, he is certainly an intriguing option for clubs to explore.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*py8d3GdDEyUCgb-rOsYubg.png" /><figcaption>Figure 8— Pavel Sulc’s Percentile Rank Performance</figcaption></figure><h4>Conclusion</h4><p>The Czech First League looks set to grow in popularity over the next few years, and it’ll be interesting to see how busy the market gets during this summer’s transfer window.</p><p>It’ll also be worth keeping an eye on whether Twenty First Group release a fresh list of the most undervalued competitions in the coming months, giving us a better idea of whether the league is still seen as a goldmine for talent. The immediate futures of the three players highlighted in this piece is also something to pay close attention to as the summer window starts to gather momentum.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=e253ec7c5f2f" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Investigating Vasteras SK’s 2024 Allsvenskan Relegation With Data]]></title>
            <link>https://medium.com/@billymulley11/investigating-vasteras-sks-2024-allsvenskan-relegation-with-data-6495289095c7?source=rss-956d94bd874b------2</link>
            <guid isPermaLink="false">https://medium.com/p/6495289095c7</guid>
            <dc:creator><![CDATA[Billy Mulley]]></dc:creator>
            <pubDate>Fri, 27 Jun 2025 08:47:08 GMT</pubDate>
            <atom:updated>2025-06-27T08:47:08.288Z</atom:updated>
            <content:encoded><![CDATA[<p>Winning Sweden’s Superettan (second-tier) in 2023, Vasteras SK started the 2024 Allsvenskan season in search of surviving the drop and establishing themselves in the country’s top-flight.</p><p>4 successive 1–0 defeats introduced Vasteras to exactly how cruel the division can be and highlighted the importance of fine margins in the higher division. The Green and White went on to finish bottom of the league, spending 21 of the 29 game weeks rooted to 16th, but this does not paint an accurate picture of how well Vasteras competed or explain why they were able to exit the division with their heads held high.</p><p>This project intends to explore open data to showcase Vasteras’ positive Allsvenskan showing, honing in on several metrics that enable us to assess their 2024 performance better than the final standings.</p><p>Straight off the bat, 13 of Vasteras’ 19 losses were by a single goal and The Green and White conceded a mere 43 goals in their 30 Allsvenskan matches. Interestingly, only the top 6 and Varnamo conceded fewer goals.</p><h4><strong>Process</strong></h4><p>In order to access high-quality and sufficient data in telling Vasteras’ story during the 2024 Allsvenskan season, I collected open data from two sources.</p><p>The majority of the data was sourced from FBref, which provided comprehensive performance data, enabling me to extract possession figures, Average Age of each squad, Goals, Shots, Shots on Target, etc., but I wanted more performance data to strengthen the story.</p><p>So, I opted to integrate data from FotMob for three additional columns: Big Chances Missed, Possession Won in the Attacking 3rd, and xG. Using Python, I was able to merge the two dataframes on ‘Club’ and create a master dataframe.</p><h4><strong>Findings</strong></h4><p>Firstly, Vasteras completed the 2024 Allsvenskan season with a comparatively younger squad than the vast majority of their divisional counterparts. The Green and White possessed the 3rd youngest squad in the division, with the average age being just 24.9.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/989/1*lfJcUnBwoFiRclUgih5oLQ.png" /><figcaption>Figure 1 — Average Age of 2024 Allsvenskan Squads</figcaption></figure><p>Now, this is where an element of subjectivity is applied to objective data, but I personally believe that teams who press high and force the issue, whilst managing lots of shots, are the most entertaining teams. As Figure 2 demonstrates, Vasteras excelled in both these metrics, suggesting they played an attacking and exciting style of football (which could certainly be backed up by the eye test). They also found themselves in very good company, with league winners Malmo and second-placed Hammarby in close proximity.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*NzyDOu9ai6tvqWoevWH9Kw.png" /><figcaption>Figure 2 — Possession Won in Attacking 3rd vs Shots</figcaption></figure><p>We can now start to pinpoint why Vasteras’ 2024 performance did not necessarily match some of the underlying metrics this project looks into. Figure 3 is a dumbbell chart that visualises Vasteras’ Expected Goals performance versus the number of Goals that they actually scored. Despite accumulating a season-long xG of 43.2 (3rd highest in the division), the Green and White managed a mere 26 and subsequently underperformed their xG by a staggering 17.2. Considering Allsvenskan is only a 30-game season, this number becomes increasingly shocking.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/989/1*M1LWvhBACmv414d-yCmJ3g.png" /><figcaption>Figure 3 — Goals vs xG: Dumbbell Chart</figcaption></figure><p>Generating the third highest xG in the division and also sitting well above the average mark for Shots and Shots on Target, we can start to hone in on Vasteras’ efficiency in front of goal as being an integral part as to why they were unable to survive the drop and secure their Allsvenskan status for another campaign. Figures 4 and 5 highlight Vasteras’ inefficiencies in front of goal further. They possessed the worst Goals per Shot on Target ratio in the division (FBref) by quite some distance and they also missed the most Big Chances (FotMob). Vasteras missed nearly 2.5 times as many Big Chances as Halmstad, who narrowly escaped the drop.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/989/1*guV0ZgLA3TCH_dhPUAYn9Q.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/989/1*vPdqFve5A4nulgehlkLCXQ.png" /><figcaption>Figure 4 — G/SoT + Big Chances Missed</figcaption></figure><p>Finally, we take a look at figure 6, which is a scatterplot that visualises the relationship between Possession and Points. I also ran a Pearson Correlation Coefficient test to measure the strength between the pair of metrics, which came out as 0.4, suggesting a moderate positive relationship. Therefore, it is really interesting to see exactly how far Vasteras stick out, considering their ball dominance but failure in accumulating enough points.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1003/1*69In2-tz6VUZ2sr1jQerZw.png" /><figcaption>Figure 5- Possession vs Points</figcaption></figure><h4><strong>Conclusion</strong></h4><p>Vasteras may have endured relegation during the 2024 Allsvenskan campaign, but it is evident why the club as a collective, and manager Kalle Karlsson, have received a lot of praise. They brought an exciting brand of football to Sweden’s top-flight and it really makes you think how they could have fared if scoring goals wasn’t such an issue.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=6495289095c7" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[‘ChatGPT, Give me a Football Scouting Mission’ — Using Data to Create a First Step in the…]]></title>
            <link>https://medium.com/@billymulley11/chatgpt-give-me-a-football-scouting-mission-using-data-to-create-a-first-step-in-the-e825cdc6785e?source=rss-956d94bd874b------2</link>
            <guid isPermaLink="false">https://medium.com/p/e825cdc6785e</guid>
            <category><![CDATA[football]]></category>
            <category><![CDATA[football-analysis]]></category>
            <category><![CDATA[football-data]]></category>
            <category><![CDATA[football-recruiting]]></category>
            <dc:creator><![CDATA[Billy Mulley]]></dc:creator>
            <pubDate>Wed, 25 Jun 2025 09:31:27 GMT</pubDate>
            <atom:updated>2025-06-25T09:31:27.912Z</atom:updated>
            <content:encoded><![CDATA[<h3>‘ChatGPT, Give me a Football Scouting Mission’ — Using Data to Create a First Step in the Recruitment Process</h3><p>Following the ‘get your work out there’ line of advice aimed at those trying to make inroads into the football recruitment industry, my latest data-driven project stems from asking ChatGPT to outline a hypothetical recruitment mission.</p><h4><strong>Mission Brief: “The Next Midfield Maestro”</strong></h4><p>ChatGPT requested a ‘central or defensive midfielder’, setting out an age range (17–22), height limit (5’10+), markets to explore (Championship, 2. Bundesliga, Liga Santander, Ligue 2, Serie B), and a target for this player to be ready for top-flight football in 2–3 years.</p><p>The beauty of this brief is that there are several ways of approaching it (from a methodology perspective) and lots of potential extensions to strengthen the process.</p><h4><strong>Defining the Role and Creating a Database</strong></h4><p>Honing in on the ambiguous ‘central or defensive midfielder’ profile provided, the first step would be to actually define what role the potential arrival would play and assess the playing philosophy of the club you are recruiting for. Sitting down and discussing this with the manager and the wider coaching team is off the cards in this hypothetical scenario, but we are going to pretend that this made-up meeting has resulted in narrowing the profile down to a highly-involved, defensively astute and impressive ball carrier who can operate as a part of a double pivot or slightly ahead as an ‘8’.</p><p>The next step of the process involved using Wyscout to collect 24/25 season event data that would fuel more advanced metrics. At this point, the filters mentioned above were applied, which produced a 62-player database to initiate the analysis. To standardise and compare values, I applied Z-scores to each of the metrics.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*gsqpi-45ZoaXdt7P6vxIOA.png" /><figcaption>Figure 1 — Original database</figcaption></figure><p>With the role defined, the database created, and with the metrics normalised, we can now start to build more advanced metrics that use composite scoring to place a higher level of importance on certain metrics. Then, the same logic is applied to create a single ‘Compatibility Score’.</p><p>It is important at this stage to be aware of the risks and limitations of using composite scores, which is perfectly detailed in an article by <a href="https://medium.com/@marclamberts/the-risks-of-using-composite-scores-in-data-scouting-90c18a17d64b">Marc Lamberts.</a> Whilst these scores enable us to rank players easily using a single value, they mask the individual elements in which the player might excel or fall short in. Equally as important, there is an immediate weighting bias, given that it is me who is ultimately selecting which metrics to give more importance to. However, composite scores can play a vital role in turning a long-list into a short-list, where other forms of analysis can then support this initial funnelling process.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Uql696WNUyawORQNISFK2A.png" /><figcaption>Figure 2 — Creation of advanced metrics</figcaption></figure><p>We are now left with a ranked list, so let’s explore a few of the names who rank highly…</p><h4><strong>Luka Hyrylainen</strong></h4><p>Top of the list, and subsequently most compatible when considering the criteria set out, is TSG Hoffenheim midfielder Luka Hyrylainen, who spent last season on loan at SSV Ulm.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*lZUmYTihc9iu65zjKlV8_A.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/745/1*viHoCLRgHJWc4aozVYuh6w.png" /><figcaption>Figure 3—Luka Hyrylainen’s performance across key metrics + percentile rank performance across advanced metrics</figcaption></figure><p>Scoring highly in the Defensive Competition and Smart Defending metrics, he ranks lowly for Ball Progression. It is evident that his strength lies in his defensive abilities and that is shown in the dashboard too, where we take a look at some of the individual metrics.</p><h4><strong>Jacob Wright</strong></h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*sClNg582F5bVqirXfGgbJA.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/745/1*NWhvz7V9PLkrsT_tF1uTrA.png" /><figcaption>Figure 4 - Jacob Wright’s performance across key metrics + percentile rank performance across advanced metrics</figcaption></figure><p>Norwich City’s Jacob Wright is next to feature, with the former Manchester City midfielder proving to be a very interesting profile. He ranks highly for both defensive advanced metrics, all whilst ranking impressively for Ball Progression. He scores very highly in Creativity, and in a different model, this may have seen him stand out even more.</p><p>As evidenced by the dashboard, he returned very impressive numbers for Defensive Duels per 90, Defensive Duels Won (%) and PAdj Interceptions. On the ball, it is his Progressive Passes per 90 that immediately stand out.</p><h4><strong>Simone Tronchin</strong></h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*TjKZkIr8zFyknx_oQCvphg.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/745/1*fzgLKUuogtIpyOHMN_Pjjw.png" /><figcaption>Figure 5— Simone Tronchin’s performance across key metrics + percentile rank performance across advanced metrics</figcaption></figure><p>Simone Tronchin’s pizza chart suggests that he is quite a well-rounded midfield option, considering the profile of midfielder we are seeking. He particularly excels in Smart Defending, but he also returns strong data in the Defensive Competition and Goal Threat metrics. He is ever so slightly below the average for Progression.</p><p>Looking at the individual metrics, he ranks 5th out of the 62 midfielders for Progressive Passes per 90, but his Progressive Runs per 90 value is much lower than the average. He also ranks highly in his Defensive Duels per 90 and PAdj Interceptions.</p><h4>Potential Next Steps/Strengthening the Process</h4><p>To strengthen this operation, there are a multitude of options for where to go from here (with time and resources). Whether that be analysing video to support or weaken our first step, using a clustering model to group similar profiles, or by integrating tracking data to provide events with more context, there are lots of avenues to explore.</p><p>Another potential enhancement would have been to factor in league quality and adjust the final ‘Compatibility Score’ accordingly.</p><p>This project also follows the basic logic that a young player thriving in the league below is a natural fit for the top tier. However, it is much more complex than that, with each team in each division having a unique game model and each team will build their role profiles slightly differently.</p><h4>Conclusion</h4><p>As Marc Lamberts’ article suggests, a model that is fueled by composite scores should not be used in isolation, which is why this particular project would act as a potential starting point within a multi-layered recruitment strategy. It is also very much from my perspective, from the outside looking in.</p><p>The purpose of this project has been to lay out a data-driven process that could initiate the recruitment process at club level, hence the lack of detail on the three players mentioned.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=e825cdc6785e" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>