<?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 Raphiel Collins on Medium]]></title>
        <description><![CDATA[Stories by Raphiel Collins on Medium]]></description>
        <link>https://medium.com/@collinsraph23?source=rss-dba1e523d653------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*ldEmsziw2sw6NIyoVZTb-g.jpeg</url>
            <title>Stories by Raphiel Collins on Medium</title>
            <link>https://medium.com/@collinsraph23?source=rss-dba1e523d653------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Sat, 21 Nov 2026 10:12:16 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@collinsraph23/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[Exploring Netflix Movies and TV Shows with Python: An Exploratory Data Analysis Project]]></title>
            <link>https://medium.com/@collinsraph23/exploring-netflix-movies-and-tv-shows-with-python-an-exploratory-data-analysis-project-b1fc75843993?source=rss-dba1e523d653------2</link>
            <guid isPermaLink="false">https://medium.com/p/b1fc75843993</guid>
            <category><![CDATA[data-visualization]]></category>
            <category><![CDATA[netflix]]></category>
            <category><![CDATA[data-analysis]]></category>
            <category><![CDATA[python-data-analysis]]></category>
            <category><![CDATA[exploratory-data-analysis]]></category>
            <dc:creator><![CDATA[Raphiel Collins]]></dc:creator>
            <pubDate>Sun, 25 Oct 2026 00:46:28 GMT</pubDate>
            <atom:updated>2026-10-25T00:46:28.473Z</atom:updated>
            <content:encoded><![CDATA[<p>How is the Netflix catalogue represented in a historical dataset? What can we learn about its mix of movies and TV shows, movie runtimes, TV seasons, genres, countries, and content ratings?</p><p>To explore these questions, I completed an exploratory data analysis (EDA) project using Python, pandas, NumPy, Matplotlib, and Seaborn.</p><p>The project took me through the data analysis workflow, from inspecting and cleaning a raw dataset to creating visualizations and communicating findings.</p><h3>Project Overview</h3><p>The objective was to investigate the structure and characteristics of a dataset containing Netflix movies and TV shows.</p><p>The analysis focused on six areas:</p><ul><li>Movies versus TV shows</li><li>Original release-year patterns</li><li>Movie runtimes and TV show seasons</li><li>Country and genre distributions</li><li>Content ratings</li><li>Catalogue additions by year and month</li></ul><p>The dataset contained <strong>8,807 records and 12 original columns</strong>.</p><p><em>Note: This is a historical dataset snapshot, not a live representation of Netflix’s current catalogue.</em></p><h3>Tools Used</h3><p>I used the following tools and libraries:</p><ul><li><strong>Python</strong> for the analysis workflow</li><li><strong>pandas</strong> for data cleaning, grouping, and aggregation</li><li><strong>NumPy</strong> for numerical operations and feature creation</li><li><strong>Matplotlib</strong> for plotting</li><li><strong>Seaborn</strong> for statistical visualizations</li><li><strong>Jupyter Notebook</strong> for documenting and organizing the analysis</li><li><strong>GitHub</strong> for version control and project presentation</li></ul><h3>Step 1: Data Profiling and Quality Checks</h3><p>Before analyzing the dataset, I inspected its dimensions, column types, unique values, missing values, and duplicate records.</p><p>The initial quality checks identified missing information in several fields:</p><p>ColumnMissing recordsDirector2,634Country831Cast825Date added10Rating4Duration3</p><p>The dataset had <strong>no exact duplicate rows and no duplicate show IDs</strong>.</p><p>These checks helped identify where cleaning was necessary and where missing information should be preserved rather than guessed.</p><h3>Step 2: Data Cleaning and Preparation</h3><p>I created a working copy of the raw data so that the original remained unchanged.</p><p>The cleaning process included:</p><ul><li>Standardizing whitespace in text fields</li><li>Labeling missing director, country, and cast information as “Not Specified”</li><li>Labeling missing ratings as “Not Rated”</li><li>Converting the date-added field to a datetime format</li><li>Separating movie runtimes from TV show season counts</li><li>Creating additional date-related columns for trend analysis</li></ul><p>I avoided inventing missing values. This was especially important for duration and date fields, where filling missing information with arbitrary values could distort the results.</p><h3>Step 3: Movies vs. TV Shows</h3><p>One of the clearest findings was the distribution of content types.</p><ul><li><strong>Movies:</strong> 6,131 titles, or 69.62%</li><li><strong>TV shows:</strong> 2,676 titles, or 30.38%</li></ul><p>Movies represented approximately seven out of every ten records in the dataset.</p><p>This gives a useful picture of the dataset’s composition. However, it does not establish that movies receive more views or perform better than TV shows, because viewing and engagement metrics were not available.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*vn5DoINaYkGzicTH0lg_-A.png" /><figcaption>Content type Distribution Chart</figcaption></figure><h3>Step 4: Movie Runtimes and TV Show Seasons</h3><p>I examined movies and TV shows separately because their duration values use different units.</p><p>The median movie runtime was <strong>98 minutes</strong>, while the mean was <strong>99.6 minutes</strong>.</p><p>The mean was slightly higher than the median, suggesting that some longer movies may pull the average upward.</p><p>For TV shows, the median recorded length was <strong>one season</strong>, compared with a mean of <strong>1.76 seasons</strong>.</p><p>These findings describe the recorded durations and seasons. They do not show whether a show was completed, cancelled, or still ongoing.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*DLG4Pf3ssg6DGgGCRz1pQA.png" /><figcaption>Movie Runtime Distribution Chart</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*WFb_TOoS-bH0CEwc0RY8kw.png" /><figcaption>TV Show Seasons Chart</figcaption></figure><h3>Step 5: Countries and Genres</h3><p>The country analysis found that the United States had the highest number of title-country mentions, with <strong>3,690 mentions</strong>.</p><p>The most frequently listed genre was <strong>International Movies</strong>, with <strong>2,752 mentions</strong>.</p><p>An important detail is that a title can list several countries and genres. Consequently, these totals are overlapping mentions rather than mutually exclusive categories.</p><p>The genre results also illustrate why field interpretation matters in data analysis. A category count should be described according to how it was calculated, rather than automatically interpreted as a count of unique productions or audience demand.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*3VhQQ__QggXhSsM0RoAzzA.png" /><figcaption>Top 10 Countries Chart</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*jbQPmzyh1bN7xMkPfmz7NA.png" /><figcaption>Top 15 Genres Chart</figcaption></figure><h3>Step 6: Content Ratings</h3><p>The most common rating label in the dataset was <strong>TV-MA</strong>, which appeared on <strong>3,207 titles</strong>.</p><p>This finding describes the distribution of recorded rating classifications. It does not tell us which ratings viewers prefer or which titles attract the most engagement.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*nQ8aYXdsh5lb5OgpYkC22w.png" /><figcaption>Content rating Distribution Chart</figcaption></figure><h3>Step 7: Catalogue Addition Trends</h3><p>I also explored the recorded dates on which titles were added to the catalogue, grouping additions by year and month.</p><p>This analysis differs from release-year analysis: a title’s original release year is not necessarily the year it was added to Netflix.</p><p>The date-added charts help describe the timing represented in the dataset. Because this is a historical snapshot, these patterns should not be presented as a complete account of Netflix’s present-day acquisition or release strategy.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*A67wGpHRiAiAG5HBHXSaRQ.png" /><figcaption>Titles added by Year Chart</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*I0ditH7gvCZXQ_6OL6oOgg.png" /><figcaption>Titles added by Month Chart</figcaption></figure><h3>What I Learned</h3><p>This project strengthened my understanding of several practical data analysis skills:</p><ol><li><strong>Data quality matters.</strong> Profiling missing values and duplicates before analysis helps prevent misleading conclusions.</li><li><strong>Cleaning decisions need context.</strong> Movie runtimes and TV seasons cannot be treated as equivalent measurements.</li><li><strong>Visualizations should answer specific questions.</strong> Each chart was built to examine a particular aspect of the catalogue.</li><li><strong>Definitions affect results.</strong> Country and genre counts may overlap, and that should be made clear when interpreting them.</li><li><strong>Limitations belong in the analysis.</strong> A dataset without viewership metrics cannot answer questions about popularity or audience behavior.</li></ol><h3>Limitations and Future Improvements</h3><p>The project is descriptive rather than predictive. It does not contain streaming views, watch time, revenue, or engagement metrics.</p><p>Potential next steps include building an interactive Tableau or Power BI dashboard, adding SQL-based analysis, and exploring additional reliable datasets that provide audience or performance measures.</p><h3>Final Thoughts</h3><p>This project gave me an opportunity to practice the complete exploratory data analysis process: inspecting a dataset, handling missing information, engineering useful features, visualizing distributions, and communicating findings with appropriate context.</p><p>The most important lesson was that good analysis involves more than generating charts. It also requires understanding what the data represents, documenting decisions, and avoiding conclusions that the data cannot support.</p><p><strong>Project repository:</strong> <a href="https://github.com/raphielcodez/netflix-eda-portfolio.git">GitHub repository link</a></p><p><strong>Dataset source:</strong> <a href="https://www.kaggle.com/datasets/shivamb/netflix-shows">Original dataset link and attribution</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=b1fc75843993" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[HR Employee Attrition Analysis Using Microsoft Excel]]></title>
            <link>https://medium.com/@collinsraph23/hr-employee-attrition-analysis-using-microsoft-excel-923547bd9c01?source=rss-dba1e523d653------2</link>
            <guid isPermaLink="false">https://medium.com/p/923547bd9c01</guid>
            <category><![CDATA[exploratory-data-analysis]]></category>
            <category><![CDATA[data-exploration]]></category>
            <category><![CDATA[excel]]></category>
            <category><![CDATA[data-analysis]]></category>
            <category><![CDATA[microsoft-excel]]></category>
            <dc:creator><![CDATA[Raphiel Collins]]></dc:creator>
            <pubDate>Thu, 22 Oct 2026 03:17:09 GMT</pubDate>
            <atom:updated>2026-10-22T03:17:09.868Z</atom:updated>
            <content:encoded><![CDATA[<p><em>An end-to-end data analytics portfolio project exploring employee retention patterns, overtime, job satisfaction, compensation, and workplace experience.</em></p><h3>Introduction</h3><p>Employee turnover can create challenges for organizations, including recruitment costs, knowledge loss, disrupted team operations, and additional pressure on remaining employees. Understanding where employee attrition is concentrated can help HR teams identify areas that deserve further investigation.</p><p>For this portfolio project, I analyzed the IBM HR Employee Attrition practice dataset using Microsoft Excel. My objective was to transform employee-level data into a structured analysis with clear metrics, comparisons, and a dashboard.</p><p>The project gave me an opportunity to apply Excel beyond basic spreadsheets by combining data preparation, formulas, data-quality checks, PivotTables, and dashboard design in one workflow.</p><h3>Project objectives</h3><p>I focused on several business questions:</p><ul><li>What proportion of employees in the dataset left the organization?</li><li>How did attrition vary across departments and job roles?</li><li>Was attrition different for employees who worked overtime?</li><li>How did job satisfaction and work-life balance relate to attrition?</li><li>How did attrition vary across monthly income and commuting-distance bands?</li><li>How did average monthly income compare between employees who left and those who stayed?</li></ul><p>These questions guided the calculations and comparisons included in the workbook.</p><h3>Tools and techniques</h3><p>The project was completed in Microsoft Excel using:</p><ul><li>Excel Tables to structure the raw and analysis-ready data</li><li>Formulas and calculated columns for additional labels and categories</li><li>Data-quality checks to review missing values, duplicate records, and consistency</li><li>PivotTables to summarize attrition across different employee groups</li><li>KPI summaries to calculate key workforce metrics</li><li>Charts and a dashboard to communicate the results visually</li></ul><p>I also prepared a data dictionary to document the fields and derived columns used in the analysis.</p><h3>Data preparation and quality checks</h3><p>The dataset contains <strong>1,470 employee records and 35 original columns</strong>.</p><p>Before interpreting the results, I checked the initial data for missing cells and exact duplicate rows. The initial checks found:</p><ul><li><strong>1,470</strong> employee records</li><li><strong>0</strong> missing cells</li><li><strong>0</strong> exact duplicate row</li><li><strong>1,470</strong> unique employee numbers</li></ul><p>I preserved the original data in the workbook and created a separate analysis-ready version. I also added readable labels for selected ordinal fields, an attrition flag, and grouped categories for income and distance from home.</p><p>These additions made the data easier to analyze and helped make the workbook more readable for someone reviewing the project.</p><p>The checks establish the condition of the dataset at the time of inspection; they do not guarantee that every field is meaningful or that the dataset represents a real workforce.</p><h3>Key findings</h3><h3>1. Overall attrition was 16.12%</h3><p>Out of 1,470 employees, 237 were recorded as having left, while 1,233 remained.</p><p>That produces an overall attrition rate of <strong>16.12%</strong>.</p><p>This metric provides a baseline for the other comparisons. Looking at the overall rate alone, however, does not show whether attrition is concentrated in particular roles or employee groups.</p><h3>2. Attrition was higher among employees who worked overtime</h3><p>The workbook showed the following group-level rates:</p><ul><li>Employees who worked overtime: <strong>30.53%</strong> attrition, with 127 departures among 416 employees.</li><li>Employees who did not work overtime: <strong>10.44%</strong> attrition, with 110 departures among 1,054 employees.</li></ul><p>The observed attrition rate among employees who worked overtime was approximately <strong>2.9 times</strong> the rate among those who did not.</p><p>This is one of the clearest differences in the analysis. It could justify further investigation into workload, scheduling, staffing levels, and employee experience. However, the data does not establish that overtime caused employees to leave, and other factors may also be involved.</p><h3>3. Attrition varied by department and job role</h3><p>At the department level, the workbook reported:</p><p>DepartmentEmployeesEmployees who leftAttrition rateSales4469220.63%Human Resources631219.05%Research &amp; Development96113313.84%</p><p>Sales had the highest department-level attrition rate among these three departments. Research &amp; Development had the largest number of departures, but it also had the largest employee population.</p><p>This illustrates why both counts and rates matter. A department can have more departures because it has more employees, even if its attrition rate is lower.</p><p>At the job-role level, Sales Representatives had an attrition rate of <strong>39.76%</strong>, with 33 departures among 83 employees. Laboratory Technicians had a rate of <strong>23.94%</strong>, with 62 departures among 259 employees.</p><p>The Sales Representative result is notable, but the group is relatively small compared with some other job roles. It should be treated as a signal for further investigation rather than a definitive conclusion about the role.</p><h3>4. Lower job satisfaction was associated with higher attrition</h3><p>The job satisfaction comparison showed:</p><ul><li>Low job satisfaction: <strong>22.84%</strong> attrition</li><li>Medium job satisfaction: <strong>16.43%</strong></li><li>High job satisfaction: <strong>16.52%</strong></li><li>Very high job satisfaction: <strong>11.33%</strong></li></ul><p>The lowest satisfaction category had the highest attrition rate, while the very-high satisfaction category had the lowest.</p><p>This pattern suggests that job satisfaction is worth exploring alongside other factors, such as management, workload, career progression, and compensation. The results do not demonstrate that satisfaction alone explains employee departures.</p><h3>5. Attrition differed across income bands</h3><p>The workbook’s income-band analysis showed a higher attrition rate in the lowest monthly-income band:</p><p>Monthly income bandEmployeesEmployees who leftAttrition rate0–3,00039511328.61%3,001–6,0005196612.72%6,001–10,0002753312.00%10,001+281258.90%</p><p>Average monthly income was approximately 4,787 among employees who left and 6,833 among employees who stayed.</p><p>These results show an association between income and attrition in this dataset. They do not prove that lower income caused departures; differences in job level, role, tenure, and other factors could influence the pattern.</p><h3>6. Work-life balance and commuting distance showed additional differences</h3><p>Employees in the lowest work-life-balance category had an attrition rate of <strong>31.25%</strong>, compared with <strong>14.22%</strong> for the “Better” category. The lowest category contained 80 employees, so its rate should be interpreted with that group size in mind.</p><p>The distance-band analysis also showed variation:</p><ul><li>0–5 miles: <strong>13.77%</strong></li><li>6–10 miles: <strong>14.47%</strong></li><li>11–15 miles: <strong>21.74%</strong></li><li>16+ miles: <strong>20.67%</strong></li></ul><p>Employees in the two longer-distance bands had higher observed attrition rates than employees in the shorter-distance bands. This may be a useful area for further exploration, but the dataset does not show that commuting distance caused turnover.</p><h3>Dashboard and workbook design</h3><p>I brought the analysis together in an Excel dashboard featuring headline metrics and charts for department attrition, attrition by job role, and average monthly income by education and attrition status.</p><p>The workbook also contains supporting sheets for the original data, cleaned data, data dictionary, data-quality checks, KPI summaries, PivotTables, and the analysis plan.</p><p>Keeping the detailed summaries alongside the dashboard allows a reviewer to move from a high-level view to the underlying group comparisons.</p><h3>Recommendations for further investigation</h3><p>Based on these descriptive patterns, a real HR analytics team could consider investigating:</p><ol><li><strong>Overtime and workload:</strong> Review overtime frequency, staffing, scheduling, and workload distribution.</li><li><strong>Role-specific retention:</strong> Explore the factors associated with higher attrition in Sales Representative and Laboratory Technician roles.</li><li><strong>Employee experience:</strong> Examine job satisfaction and work-life balance alongside other workplace measures.</li><li><strong>Compensation and career progression:</strong> Assess income patterns in the context of role, level, tenure, and promotion opportunities.</li><li><strong>Commuting challenges:</strong> Investigate whether travel demands or commuting arrangements are relevant to the observed distance patterns.</li></ol><p>These are potential directions for investigation, not prescriptions based on proven causes. Any decisions should use appropriate context, additional evidence, and fair treatment of employees.</p><h3>Limitations</h3><p>This project uses a fictional, observational practice dataset. The analysis is descriptive and does not establish causation. Some groups are much smaller than others, which can make their attrition rates less stable. The analysis also does not control for all possible confounding factors or build a predictive model.</p><p>The results should therefore be understood as an exploration of patterns within this dataset, not as a diagnosis of a real employer or an employee-level risk assessment.</p><h3>What I learned</h3><p>This project helped me practice structuring a data-analysis workflow in Excel, checking data quality, creating analysis-friendly fields, summarizing group-level metrics, and communicating results through a dashboard.</p><p>It also reinforced an important analytical lesson: counts, rates, and group sizes tell different parts of the story. A useful dashboard should help people ask better questions, not encourage them to jump from a correlation to a causal explanation.</p><h3>Conclusion</h3><p>This project demonstrates how Microsoft Excel can support a complete introductory HR analytics workflow, from data preparation and validation to KPI reporting and dashboard presentation.</p><p>The strongest observed differences in this dataset involved overtime status, job role, job satisfaction, income band, and work-life balance. Each offers a potential starting point for deeper analysis, provided the findings are interpreted carefully and in context.</p><p><strong>Project files:</strong> <a href="https://github.com/raphielcodez/ibm-hr-employee-attrition-excel.git">GitHub repository link</a></p><p><strong>Dashboard:</strong> See the dashboard screenshot included in the repository.</p><p>Thank you for reading. I’m continuing to build my data analytics portfolio and strengthen my ability to turn raw data into clear, responsible, and useful insights.</p><p>#DataAnalytics #Excel #HRAnalytics #DataVisualization #PortfolioProject</p><p>#DataAnalytics #Excel #HRAnalytics #DataVisualization #PortfolioProject</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=923547bd9c01" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Analyzing Adani Group Stock Performance Using Python, SQL, and Tableau]]></title>
            <link>https://medium.com/@collinsraph23/analyzing-adani-group-stock-performance-using-python-sql-and-tableau-d0a9ff08ced0?source=rss-dba1e523d653------2</link>
            <guid isPermaLink="false">https://medium.com/p/d0a9ff08ced0</guid>
            <category><![CDATA[python-data-analysis]]></category>
            <category><![CDATA[stocks]]></category>
            <category><![CDATA[data-analysis]]></category>
            <category><![CDATA[data-analysis-tool]]></category>
            <category><![CDATA[adani-group]]></category>
            <dc:creator><![CDATA[Raphiel Collins]]></dc:creator>
            <pubDate>Tue, 06 Oct 2026 23:57:16 GMT</pubDate>
            <atom:updated>2026-10-07T03:52:25.298Z</atom:updated>
            <content:encoded><![CDATA[<p><em>An end-to-end data analytics project exploring historical stock performance, volatility, trading activity, and market behavior.</em></p><h3>Introduction</h3><p>Financial markets generate enormous amounts of data every trading day.</p><p>Opening prices, closing prices, trading volume, daily highs and lows, dividends, and stock splits can provide valuable information — but raw financial data by itself doesn’t tell a complete story.</p><p>For this project, I wanted to take a historical stock-market dataset and turn it into something more meaningful: an analytical workflow that could identify trends, compare companies, investigate volatility, and communicate findings visually.</p><p>I chose historical stock data covering companies within the <strong>Adani Group</strong> and built an end-to-end analysis using three core tools:</p><ul><li><strong>Python</strong> for data cleaning, transformation, and exploratory analysis</li><li><strong>SQL</strong> for structured analysis and ranking</li><li><strong>Tableau</strong> for interactive dashboards and visualization</li></ul><p>The project ultimately became a complete pipeline:</p><pre>Raw Data<br>   ↓<br>Python<br>   ↓<br>Data Cleaning &amp; Feature Engineering<br>   ↓<br>SQL Analysis<br>   ↓<br>Tableau Dashboards<br>   ↓<br>Insights &amp; Communication</pre><h3>The Questions I Wanted to Answer</h3><p>Rather than simply creating charts from the dataset, I wanted the analysis to answer specific questions.</p><h3>Performance</h3><ul><li>How have the companies performed over time?</li><li>Which companies experienced the strongest cumulative growth?</li><li>How different was performance between companies?</li></ul><h3>Risk and volatility</h3><ul><li>Which companies experienced the greatest price volatility?</li><li>How frequently did companies experience positive trading days?</li><li>Were periods of large price movements associated with increased volatility?</li></ul><h3>Trading activity</h3><ul><li>Which companies had the highest average trading volume?</li><li>When did trading activity increase?</li><li>Did large price movements coincide with unusual trading activity?</li></ul><h3>January 2023</h3><p>I also wanted to investigate stock behavior surrounding <strong>January 24, 2023</strong>.</p><p>Rather than treating this date as proof of causation, I used it as an event date and examined the surrounding period to see whether there were noticeable changes in returns, volatility, and trading activity.</p><h3>Understanding the Dataset</h3><p>The dataset contains historical daily stock-market observations for companies within the Adani Group.</p><p>The main fields included:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*mSNuDKZ75KEVwwxzCxAuHw.jpeg" /></figure><p>The dataset contained approximately <strong>40,000+ observations</strong> and <strong>11 unique company symbols</strong> across a long historical period.</p><p>One important issue emerged during the initial exploration: not every company had the same amount of historical data.</p><p>For example, <strong>Sanghi Industries had only a single observation</strong>, making it unsuitable for meaningful long-term performance comparisons.</p><p>This was an early reminder that data analysis isn’t just about calculating metrics. Before comparing groups, you need to understand whether the underlying data actually supports the comparison.</p><h3>Step 1: Cleaning the Data with Python</h3><p>I started the analysis in Python using Pandas, NumPy, Matplotlib, and Seaborn.</p><p>The first stage involved inspecting the dataset for:</p><ul><li>Missing values</li><li>Duplicate observations</li><li>Incorrect data types</li><li>Date formatting issues</li><li>Company coverage</li><li>Inconsistent historical coverage</li></ul><p>The original timestamp was converted into a proper datetime field:</p><pre>df[&quot;date&quot;] = pd.to_datetime(<br>    df[&quot;timestamp&quot;],<br>    unit=&quot;s&quot;,<br>    utc=True<br>).dt.tz_convert(None)</pre><p>I then sorted the observations by company and date:</p><pre>df = df.sort_values(<br>    [&quot;symbol&quot;, &quot;date&quot;]<br>).reset_index(drop=True)</pre><p>Duplicate company-date observations were removed:</p><pre>df = df.drop_duplicates(<br>    subset=[&quot;symbol&quot;, &quot;date&quot;]<br>).reset_index(drop=True)</pre><p>The cleaned dataset was then saved for use in the SQL and Tableau portions of the project.</p><h3>Step 2: Creating New Analytical Features</h3><p>The raw dataset contained prices and volume, but I wanted to derive metrics that would allow me to analyze performance and risk.</p><h3>Daily Returns</h3><p>One of the most important metrics was daily return.</p><p>Daily return measures the percentage change in closing price between consecutive trading days.</p><pre>df[&quot;daily_return&quot;] = (<br>    df.groupby(&quot;symbol&quot;)[&quot;close&quot;]<br>      .pct_change(fill_method=None)<br>)</pre><p>This gave me a much more useful metric for comparing companies than simply looking at their raw stock prices.</p><p>A company trading at $20 and another trading at $2,000 can’t be meaningfully compared based solely on their prices.</p><p>Percentage returns provide a common basis for comparison.</p><h3>Intraday Return</h3><p>I also calculated the change between the opening and closing price:</p><pre>df[&quot;intraday_return&quot;] = (<br>    (df[&quot;close&quot;] - df[&quot;open&quot;]) / df[&quot;open&quot;]<br>)</pre><p>This helps identify whether a stock generally finished the trading period above or below its opening price.</p><h3>Daily Price Range</h3><p>To measure the amount of price movement within a trading day, I calculated:</p><pre>df[&quot;daily_range_pct&quot;] = (<br>    (df[&quot;high&quot;] - df[&quot;low&quot;]) / df[&quot;open&quot;]<br>)</pre><p>This provides another way of examining market activity and price movement.</p><h3>Volume Change</h3><p>Trading volume can provide useful context when looking at price movements.</p><p>I calculated the percentage change in volume between consecutive observations:</p><pre>df[&quot;volume_change_pct&quot;] = (<br>    df.groupby(&quot;symbol&quot;)[&quot;volume&quot;]<br>      .pct_change(fill_method=None)<br>)</pre><p>This allowed me to investigate whether periods of significant price movement were accompanied by changes in trading activity.</p><h3>Step 3: Measuring Volatility</h3><p>Price performance tells only part of the story.</p><p>Two companies could have similar returns while having completely different levels of risk.</p><p>To examine this, I calculated annualized volatility based on daily returns.</p><p>I also created a 30-day rolling volatility measure:</p><pre>df[&quot;rolling_volatility_30d&quot;] = (<br>    df.groupby(&quot;symbol&quot;)[&quot;daily_return&quot;]<br>      .transform(<br>          lambda x: x.rolling(<br>              window=30,<br>              min_periods=20<br>          ).std() * np.sqrt(252)<br>      )<br>)</pre><p>The use of 252 reflects the approximate number of trading days in a year.</p><p>This allowed me to observe how volatility changed over time rather than relying on a single overall figure.</p><h3>Step 4: Comparing Cumulative Performance</h3><p>To compare how companies performed over their available historical periods, I calculated cumulative return.</p><pre>df[&quot;cumulative_return&quot;] = (<br>    df.groupby(&quot;symbol&quot;)[&quot;close&quot;]<br>      .transform(<br>          lambda x:<br>          x / x.dropna().iloc[0] - 1<br>      )<br>)</pre><p>I also created a normalized price index, where each company’s starting price was represented as 100.</p><p>This makes it much easier to compare companies that have very different absolute stock prices.</p><p>Instead of asking:</p><blockquote><em>“Which stock has the highest price?”</em></blockquote><p>I could ask:</p><blockquote><em>“Which stock experienced the greatest relative growth over its available history?”</em></blockquote><h3>Step 5: Moving to SQL</h3><p>After cleaning and exploring the data with Python, I moved to SQL.</p><p>This was an important part of the project because I wanted to demonstrate that the analysis wasn’t dependent on a single programming language.</p><p>Python was useful for data preparation and exploratory analysis.</p><p>SQL allowed me to approach the same dataset from a structured analytical perspective.</p><p>The SQL analysis covered:</p><ul><li>Dataset profiling</li><li>Data-quality checks</li><li>Duplicate detection</li><li>Price statistics</li><li>Daily price ranges</li><li>Trading volume</li><li>Corporate actions</li><li>Daily returns</li><li>Positive and negative trading days</li><li>Volatility</li><li>Extreme price movements</li><li>Yearly performance</li><li>Monthly trading volume</li><li>Company rankings</li><li>January 2023 event analysis</li></ul><h3>Ranking Companies with SQL</h3><p>One of the more useful SQL techniques in the project was the use of window functions.</p><p>For example, I used ranking logic to identify companies with the largest daily movements and highest trading activity.</p><p>This is useful because it turns a large dataset into targeted questions such as:</p><blockquote><em>Which companies experienced the largest positive daily movements?</em></blockquote><p>or:</p><blockquote><em>What were the highest-volume trading days for each company?</em></blockquote><p>Rather than manually inspecting thousands of rows, SQL allows these questions to be answered systematically.</p><h3>Step 6: Investigating January 24, 2023</h3><p>One of the more interesting parts of the project was the event analysis surrounding <strong>January 24, 2023</strong>.</p><p>Instead of looking only at the event date itself, I created a window around the date.</p><p>The Python analysis used approximately <strong>45 days before and after the selected event date</strong>.</p><pre>event_date = pd.Timestamp(&quot;2023-01-24&quot;)</pre><pre>event_df = df[<br>    (df[&quot;date&quot;] &gt;= event_date - pd.Timedelta(days=45))<br>    &amp;<br>    (df[&quot;date&quot;] &lt;= event_date + pd.Timedelta(days=45))<br>].copy()</pre><p>I then divided the observations into:</p><ul><li>Before the event</li><li>After the event</li></ul><p>and examined:</p><ul><li>Average daily return</li><li>Return volatility</li><li>Average trading volume</li><li>Large price movements</li></ul><p>This approach provided a more useful view than looking at one trading day in isolation.</p><h3>Why I Avoided Claiming Causation</h3><p>One of the most important lessons from this analysis was the difference between <strong>correlation and causation</strong>.</p><p>If stock prices or trading volume changed around January 24, 2023, that doesn’t automatically mean that the selected event caused those changes.</p><p>Stock prices can be affected by many factors:</p><ul><li>Company announcements</li><li>Investor sentiment</li><li>Broader market movements</li><li>Economic conditions</li><li>Industry developments</li><li>News events</li><li>Trading activity</li><li>Other external factors</li></ul><p>Therefore, this project treats January 24 as an <strong>event date for observational analysis</strong>, not as proof of causality.</p><p>This distinction is important when working with financial data.</p><h3>Step 7: Building the Tableau Dashboards</h3><p>Once the Python and SQL analysis was complete, I moved the cleaned dataset into Tableau.</p><p>The goal wasn’t simply to reproduce the Python charts.</p><p>Instead, I wanted to create an interactive analytical experience where someone could explore the companies and time periods themselves.</p><p>I created three main dashboards.</p><h3>Dashboard 1: Executive Overview</h3><p>The Executive Overview provides a high-level view of the dataset.</p><p>It includes:</p><ul><li>Closing-price trends</li><li>Daily returns</li><li>Trading volume</li><li>Company filters</li><li>Date filters</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*GDOHA2UQl6NL_XMSw5hPRw.png" /><figcaption><em>Executive Overview dashboard created in Tableau.</em></figcaption></figure><p>The dashboard allows users to select individual companies and examine how price, returns, and trading activity changed over time.</p><h3>Dashboard 2: Company Performance</h3><p>The second dashboard focuses on comparing companies.</p><p>It includes:</p><ul><li>Cumulative return</li><li>Annualized volatility</li><li>Positive trading-day percentage</li><li>Average daily trading volume</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*j7xrp2geJ-Gdce4n24djSg.png" /><figcaption><em>Company Performance dashboard created in Tableau.</em></figcaption></figure><p>I wanted this dashboard to demonstrate why looking at only one metric can be misleading.</p><p>For example, a company may have strong cumulative performance but also exhibit significantly higher volatility.</p><p>Similarly, a company with a high percentage of positive trading days isn’t necessarily the company with the highest overall return.</p><p>Looking at multiple metrics together provides a more complete picture.</p><h3>Dashboard 3: January 2023 Event Analysis</h3><p>The third dashboard focuses specifically on the period surrounding January 24, 2023.</p><p>It combines:</p><ul><li>Closing-price movement</li><li>Daily returns</li><li>Trading volume</li><li>Company-level comparisons</li><li>Event-date visualization</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*MbdkEYpAMMXjrKNJt2r0vg.png" /><figcaption><em>January 2023 Event Analysis dashboard created in Tableau.</em></figcaption></figure><p>A visual marker identifies January 24, 2023, allowing the viewer to compare the period before and after the selected event date.</p><p>This makes the event-window analysis much easier to interpret than looking at a table of numbers alone.</p><h3>What I Learned from the Analysis</h3><p>One of the biggest takeaways from this project was that <strong>stock performance cannot be evaluated using a single metric</strong>.</p><p>A meaningful comparison requires looking at several dimensions simultaneously:</p><h3>Return</h3><p>How much did the stock’s value change?</p><h3>Volatility</h3><p>How much did the stock’s daily returns fluctuate?</p><h3>Trading Volume</h3><p>How active was trading?</p><h3>Positive Trading Days</h3><p>How frequently did the stock record a positive daily return?</p><h3>Time</h3><p>Did the behavior change during particular periods?</p><p>Combining these metrics provides a more complete analytical picture.</p><h3>What I Learned as a Data Analyst</h3><p>This project was also valuable because it forced me to work through an entire analytics workflow rather than focusing on a single tool.</p><p>I learned that each tool has a different role.</p><h3>Python</h3><p>Python was useful for:</p><ul><li>Cleaning data</li><li>Transforming data</li><li>Creating new metrics</li><li>Exploratory analysis</li><li>Statistical calculations</li></ul><h3>SQL</h3><p>SQL helped me think about the data differently.</p><p>Instead of manipulating the dataset primarily through Python, I had to formulate analytical questions using:</p><ul><li>Aggregations</li><li>Grouping</li><li>Window functions</li><li>Rankings</li><li>Time-based queries</li></ul><h3>Tableau</h3><p>Tableau taught me the importance of communication.</p><p>A technically correct analysis isn’t automatically useful if the audience cannot understand it.</p><p>The dashboards forced me to think about:</p><ul><li>Which metrics matter?</li><li>Which comparisons are useful?</li><li>How should information be arranged?</li><li>What should the user be able to filter?</li><li>How can complex analysis be communicated quickly?</li></ul><h3>Challenges</h3><p>The project also came with several challenges.</p><h3>Unequal historical coverage</h3><p>Not every company had the same amount of historical data.</p><p>This meant I needed to be careful when comparing long-term performance.</p><h3>Missing values</h3><p>There were missing closing-price observations in the original dataset.</p><p>These had to be considered when calculating returns and other derived metrics.</p><h3>Financial metrics</h3><p>Metrics such as volatility and cumulative return required careful interpretation.</p><p>A metric can be mathematically correct while still being misleading if the underlying data or context isn’t considered.</p><h3>Event analysis</h3><p>The January 2023 analysis required particular caution because observing a change around an event does not establish causation.</p><p>These challenges were useful because they reinforced an important principle of data analysis:</p><blockquote><strong><em>Good analysis isn’t just about producing a result. It’s about understanding how trustworthy and meaningful that result actually is.</em></strong></blockquote><h3>Limitations</h3><p>There are several limitations to this project.</p><p>First, the analysis is based on historical data and therefore should not be interpreted as a prediction of future stock performance.</p><p>Second, companies have different historical coverage, which can affect long-term comparisons.</p><p>Third, the January 2023 event analysis is observational. It does not establish that the selected event caused changes in stock prices or trading volume.</p><p>Finally, the analysis does not incorporate every possible factor affecting stock prices, such as macroeconomic conditions, broader market indices, investor sentiment, or other company-specific events.</p><p>This project is intended as a <strong>data analytics portfolio project</strong>, not financial advice or an investment recommendation.</p><h3>Future Improvements</h3><p>There are several ways I could extend this project.</p><h3>1. Add a market benchmark</h3><p>Comparing the companies against an index such as the <strong>NIFTY 50</strong> would provide additional context.</p><p>Instead of asking only:</p><blockquote><em>“Did the company increase?”</em></blockquote><p>I could ask:</p><blockquote><em>“Did the company outperform the broader market?”</em></blockquote><h3>2. Add maximum drawdown</h3><p>Maximum drawdown would help measure the largest decline from a previous peak.</p><h3>3. Calculate Sharpe ratios</h3><p>Risk-adjusted performance would provide another way of comparing companies.</p><h3>4. Perform a formal event study</h3><p>The January 2023 analysis could be expanded into a more formal event-study methodology, including expected returns and statistical significance testing.</p><h3>5. Automate the data pipeline</h3><p>A future version could automatically retrieve new market data, process it, update the database, and refresh the Tableau dashboard.</p><h3>Final Thoughts</h3><p>This project started with a simple question:</p><blockquote><strong><em>What can I actually learn from historical stock-market data?</em></strong></blockquote><p>It turned into a much broader exercise in data analysis.</p><p>I learned how to take a relatively messy dataset, clean it, transform it into useful analytical features, analyze it using SQL, and finally communicate the results through interactive dashboards.</p><p>More importantly, I learned that being a data analyst isn’t simply about knowing Python, SQL, or Tableau individually.</p><p>The real value comes from connecting the pieces:</p><pre>Raw Data<br>   ↓<br>Clean Data<br>   ↓<br>Good Questions<br>   ↓<br>Analysis<br>   ↓<br>Visualization<br>   ↓<br>Insights<br>   ↓<br>Communication</pre><p>The tools are important, but the ability to ask meaningful questions and communicate what the data is telling you is just as important.</p><h3>🔗 Project Resources</h3><p>The complete project, including the Python notebook, SQL analysis, cleaned dataset, and project documentation, is available on GitHub.</p><p><strong>GitHub:</strong> <a href="https://github.com/raphielcodez/adani-stock-analysis">https://github.com/raphielcodez/adani-stock-analysis</a></p><p><strong>Tableau Public:</strong> <a href="https://public.tableau.com/app/profile/raphiel.collins/vizzes">https://public.tableau.com/app/profile/raphiel.collins/vizzes</a></p><h3>Tools Used</h3><p><strong>Python • Pandas • NumPy • Matplotlib • Seaborn • SQL • Tableau • GitHub</strong></p><p><em>This project was created as part of my data analytics portfolio to demonstrate an end-to-end workflow for transforming raw data into structured analysis, interactive visualization, and actionable insights.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=d0a9ff08ced0" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>