<?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 Anil Celik on Medium]]></title>
        <description><![CDATA[Stories by Anil Celik on Medium]]></description>
        <link>https://medium.com/@acelik?source=rss-3d9e2e9fd419------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*KSpb1-iz7CF9TAtIMvWErg.jpeg</url>
            <title>Stories by Anil Celik on Medium</title>
            <link>https://medium.com/@acelik?source=rss-3d9e2e9fd419------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Mon, 16 Nov 2026 13:57:29 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@acelik/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[Story of Chestnut]]></title>
            <link>https://medium.com/chestnut-insurance/story-of-chestnut-14a0d18988f2?source=rss-3d9e2e9fd419------2</link>
            <guid isPermaLink="false">https://medium.com/p/14a0d18988f2</guid>
            <category><![CDATA[insurance]]></category>
            <category><![CDATA[environment]]></category>
            <category><![CDATA[california]]></category>
            <category><![CDATA[wildfires]]></category>
            <dc:creator><![CDATA[Anil Celik]]></dc:creator>
            <pubDate>Fri, 01 Nov 2019 23:21:39 GMT</pubDate>
            <atom:updated>2019-11-01T23:21:39.746Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*of3kv3xsv0vBPx0iAxIhXw.png" /><figcaption>Chestnut Homepage — getchestnut.com</figcaption></figure><p>Five years ago, my co-founder (<a href="https://www.linkedin.com/in/nilgun-celik-6052651a/">Nilgun Dag</a>) and I started a company called UrbanStat.</p><p><a href="http://www.urbanstat.com/">UrbanStat</a> helps insurance companies to understand catastrophic or human-made events better using location intelligence, visualization, and machine learning algorithms. Companies like Sompo Japan, Ageas, Allianz, Safety Insurance, and many more are using our platform today to manage their portfolios, underwrite their customers, and design better products for people and businesses.</p><figure><img alt="Portfolio analysis during an actual hurricane with real-time data analysis" src="https://cdn-images-1.medium.com/max/1024/0*dw5ywZek0ucEldG2.jpg" /><figcaption>Portfolio analysis during an actual hurricane with real-time weather data</figcaption></figure><p>Since early 2018, we have been receiving questions about wildfires from insurance carriers. In the past, wildfires were still occuring, but they were not impacting the insured properties to the effect that they are today. Wildfires were not a major problem for insurance companies until 2017, which was named the most destructive wildfire season in California. 9,133 fires burned 1,381,405 acres and destroyed or damaged 10,000 structures. Needless to say, the industry didn’t really care about this problem until now. The wildfire phenomenon is something they don’t understand yet. This is why they don’t know how to approach the problem.</p><p>On December 2018, Joyce Howard, a broker in Auburn explained:</p><blockquote>All these major companies started pulling out quietly. People got non-renewals; people got flat-out canceled. There are companies that are still doing that today.</blockquote><p>On May 2019, Christopher Kerksieck, a Farmer Insurance Agency Owner stated:</p><blockquote>Generally, it’s two to three times higher than the premium that they were paying before</blockquote><p>explaining premium surcharges on homeowners / fire insurance.</p><p>On May 2019, Becca Habegger reported on ABC10:</p><blockquote>Some Californians trying to get ready for fire season by buying fire insurance are finding that they can’t — because their carrier isn’t offering it to them anymore.</blockquote><p>Newspapers, magazines, news sites, and social media are full of similar experiences. Californians are facing a major insurance crisis which is expected to only get worse.</p><p>With the intention to help insurance carriers, and indirectly help the people, we have decided to implement a new wildfire risk map for California. After a year long R&amp;D, we have released our wildfire map for California during the Spring of 2019. The problem with current maps is that they are based on simple statistical methodologies, limited variables (often the unnecessary ones), and sometimes rule based decision processes. Let me explain this in more detail:</p><p>Wildfires can start from any number of reasons. Arson, lightning strikes, failing power infrastructure (e.g. Camp Fire), shattered glasses, forgotten campfires and so on. Once a fire starts, it can accelerate based on any number of factors: vegetation (often referred as fuel), temperature, winds, precipitation, humidity, slope (topography of the area) and so on.</p><p>When we look at current approaches, we immediately see that a limited number of factors are being incorporated to the wildfire maps/models:</p><ul><li>Slope</li><li>Vegetation (Fuel, bush etc.)</li><li>Access to roads</li><li>Access to fire stations</li></ul><p>However, our research shows that we can easily include over 20+ variables to modeling process.</p><p>We have sourced the data for the last 80 years of wildfires in California. We also have sourced over 20 different variables that could explain why wildfires are starting in a specific area. Our model has used the first 60 years of wildfires to learn and validate (training/validation), and the last 20 years of wildfires to test its ability to predict unseen data (blind test). We have carefully validated our model using several validation techniques, but I won’t go into details on this post.</p><p>The result of this process is the map below:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1002/1*Dyx-1w1CfjoRlCuSnag1kA.jpeg" /><figcaption>UrbanStat California Wildfire Risk Map — All Rights Reserved</figcaption></figure><p>This map is nothing but a fancy visualization unless the numbers behind it are explained. In order to evaluate the performance of this map we have compared it to only publicly available map that is provided by US Forestry / USGS. Both US Forestry and UrbanStat have classified around 35% of California as High or higher risk when it comes to wildfires. In UrbanStat’s case, High, Very High, and Extreme makes up around 36% of California.</p><p><strong>Our Evaluation</strong></p><p>The simplest thing we could ask ourselves was how this map performed on a data it has never seen (which is the last 20 years of wildfires), and we did the same evaluation against US Forestry’s map where they actually used the last 20 years of data to improve their map.</p><p>What percentage of the areas that burned in last 20 years was classified as High or higher risk when it comes to wildfires?</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/991/1*J5CWjXNaMnZatPaFXJw02g.png" /></figure><p>This analysis proves that UrbanStat’s map was almost 50% better than the US Forestry’s map when it comes to explaining the wildfires.</p><p>According to UrbanStat’s map, Camp, Saddleridge, and Kincade fires all started either in High, Very High, or Extreme risk areas and they primarily spread through similar areas.</p><p><strong>What’s Chestnut?</strong></p><p>Chestnut is our response to the crisis in California. We have created this product due to the unwillingness of insurance companies to solve this financial crisis. We picked the name Chestnut because the American Chestnut tree has <a href="https://getchestnut.com/story-of-american-chestnut/">an amazing story</a> about loss, adaption, and perseverance, just like the victims of wildfires.</p><p>Chestnut is better at understanding the risk of wildfire and therefore it is better at pricing the risk for insurance providers. UrbanStat has been helping insurance carriers around the world to manage their portfolio better for years now. UrbanStat’s ability to underwrite and price better will enable Chestnut to provide a much needed financial relief to victims of wildfires. <strong>More than half of the claims are still open (unpaid) after the wildfires in 2018.</strong></p><p>Our plan is subject to change, but our goal is to implement a financial product that will pay up to $20,000 within days after a wildfire. We plan to cover following damages:</p><ul><li>Damages to your house, personal property, vehicles, landscaping, and other structures</li><li>$100/day you are away from home due to forced evacuations up to 20 days a year</li><li>$100/day you spend in a hospital up to 20 days a year</li><li>$50/day your home is out of power up to 10 days a year</li></ul><p><strong>Chestnut is not a replacement for your home insurance. </strong>However, it will help you reduce your home insurance premium and, in certain cases, it may be the only solution that you have access to, with home insurance plans proving difficult to get in California every day. We are taking the necessary steps to ensure that Chestnut complies with all regulatory agencies. We appreciate your patience as our product needs time to be released fully. Insurance is a highly regulated industry and we are working relentlessly to get this ready before the next wildfire season. Until then, we’ve decided to open our risk map to public, so you can <strong>analyze your property’s wildfire risk now.</strong></p><p>For free wildfire risk analysis: <a href="http://analyze.getchestnut.com/">analyze.getchestnut.com</a></p><p>For more information: <a href="http://getchestnut.com/">getchestnut.com</a></p><p>Anil Celik</p><p>Co-Founder &amp; CEO at <a href="http://www.urbanstat.com/">UrbanStat</a> and <a href="https://www.getchestnut.com/">Chestnut</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=14a0d18988f2" width="1" height="1" alt=""><hr><p><a href="https://medium.com/chestnut-insurance/story-of-chestnut-14a0d18988f2">Story of Chestnut</a> was originally published in <a href="https://medium.com/chestnut-insurance">Chestnut</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[California Fires Break Out Overnight: Kincade Fire & Tick Fire]]></title>
            <link>https://medium.com/chestnut-insurance/california-fires-break-out-overnight-kincade-fire-tick-fire-a7fce70a1f99?source=rss-3d9e2e9fd419------2</link>
            <guid isPermaLink="false">https://medium.com/p/a7fce70a1f99</guid>
            <category><![CDATA[environment]]></category>
            <dc:creator><![CDATA[Anil Celik]]></dc:creator>
            <pubDate>Thu, 31 Oct 2019 18:21:03 GMT</pubDate>
            <atom:updated>2019-10-31T18:21:03.705Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*EYUElhl4HzsN1fybDl2Daw.jpeg" /></figure><p>As California residents anticipated the end of the wildfire season, two wildfires fires broke out on Thursday. The Kincade fire is quickly burning through Sonoma County and the Tick fire is impacting Los Angeles.</p><p>Tick Fire broke out early afternoon on Thursday. The fire is a wind-driven brush fire that started on the 31600 block of Tick Canyon Road. The cause of the fire is still unknown. Early Friday morning, the brush fire jumped the 14 freeway and caused the evacuation of 40,000 residents and threatening 10,000 homes. Tick fire has burned 4,000 acres with 5% contained.</p><p>Thursday evening, families were forced to evacuate their homes in the dark after a fire broke out in a remote and sparsely populated area of Geyserville around 9pm. As of early Friday morning, the Kincade fire has burned through 16,000 acres and is 5% contained. Although officials have not confirmed the source of the fire, PG&amp;E reported that one of their transmission towers was still powered when the fire broke out. Late Thursday, PG&amp;E cut power to thousands of homes amid dangerous weather conditions to prevent a wildfire outbreak. PG&amp;E claimed that they became aware that a transmission tower was still powered around 9:20pm in Geyserville and the Kincade fire broke out at 9:27pm. This would not be the first time that a faulty transmission tower caused a major wildfire. The deadly Camp Fire of 2018 was caused by PG&amp;E transmission lines.</p><p>Other, smaller, fires have broken out overnight according to the California Fire Department. Most have remainder under 100 acres and are 25–50% contained. Strong winds are expected to pick up Friday evening and into the weekend, making the Kincade and Tick fires difficult to manage. As firefighters work overnight to protect homes, residents are reminded to say alert and evacuate at a moment’s notice.</p><p>Check <a href="http://getchestnut.com/">getchestnut.com</a> to get your free wildfire risk analysis.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=a7fce70a1f99" width="1" height="1" alt=""><hr><p><a href="https://medium.com/chestnut-insurance/california-fires-break-out-overnight-kincade-fire-tick-fire-a7fce70a1f99">California Fires Break Out Overnight: Kincade Fire &amp; Tick Fire</a> was originally published in <a href="https://medium.com/chestnut-insurance">Chestnut</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[How Wildfires Impact Insurance Policies]]></title>
            <link>https://medium.com/chestnut-insurance/how-wildfires-impact-insurance-policies-b4299c08064f?source=rss-3d9e2e9fd419------2</link>
            <guid isPermaLink="false">https://medium.com/p/b4299c08064f</guid>
            <category><![CDATA[insurance]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[california]]></category>
            <category><![CDATA[wildfires]]></category>
            <dc:creator><![CDATA[Anil Celik]]></dc:creator>
            <pubDate>Wed, 30 Oct 2019 17:15:05 GMT</pubDate>
            <atom:updated>2019-10-30T17:15:05.575Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*sOpYo0TA3P-GF_Wc7kYYTQ.jpeg" /><figcaption>A View from Chestnut’s Wildfire Risk Map</figcaption></figure><p>California residents are struggling to rebuild after wildfire damage. Over the last few years, residents claimed that insurance companies are pulling back their plans in areas where wildfire activity has increased.</p><p>Traditionally, insurance providers would survey the land and assess the potential risk of each home. Today, insurance providers are denying insurance renewals in areas where there has been a fire, even if the home up for renewal has not made a claim. Additionally, homeowners are having trouble finding plans from providers or their current rate has skyrocketed. A resident of Squaw Valley reported that her homeowners insurance rate jumped from $900 per year to $6,200 per year.</p><p>Residents who cannot find coverage on the voluntary market are forced to find coverage under the FAIR Plan at a much higher cost. Between 2015 and 2018, the number of people who enrolled in FAIR Plan policies grew my 177%.</p><p>Homeowners who file a claim with their insurance provider can wait to receive their money for up to 2 years. After the devastating Paradise Park fires, residents found that their payout delayed and insurance companies did not have the necessary funds to pay the cost of damage in full. In some areas, state laws require insurers to pay out the claims within 10–30 days, while some ask for insurance companies to pay out in a “reasonable amount of time.” In California, insurance companies have 40 days to either accept or deny a claim with the ability to request more time and continue to update the policy owner every 30 days.</p><p>What does this mean for homeowners in California? Homeowners should expect that their rates may increase or that they may be denied insurance. Residents should <a href="http://getchestnut.com/">check</a> if they are in an area at risk of wildfires and assess their current insurance policy to make sure that they have the best coverage.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=b4299c08064f" width="1" height="1" alt=""><hr><p><a href="https://medium.com/chestnut-insurance/how-wildfires-impact-insurance-policies-b4299c08064f">How Wildfires Impact Insurance Policies</a> was originally published in <a href="https://medium.com/chestnut-insurance">Chestnut</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The Future of Wildfires]]></title>
            <link>https://medium.com/chestnut-insurance/the-future-of-wildfires-f5544f162a8e?source=rss-3d9e2e9fd419------2</link>
            <guid isPermaLink="false">https://medium.com/p/f5544f162a8e</guid>
            <category><![CDATA[wildfires]]></category>
            <category><![CDATA[insurance]]></category>
            <category><![CDATA[climate-change]]></category>
            <category><![CDATA[california]]></category>
            <dc:creator><![CDATA[Anil Celik]]></dc:creator>
            <pubDate>Wed, 30 Oct 2019 17:13:37 GMT</pubDate>
            <atom:updated>2019-10-31T18:00:11.641Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*5JdlypLfAbHynVKDcP5DYg.jpeg" /></figure><p>The recent wildfires in California have sparked concern with homeowners and renters in fire prone areas. With wetter winters and drier summers, the California landscape becomes the perfect fuel for wildfires. To understand property risk, we need to assess the past, present and future of wildfires and how climate change will help us predict how to prepare for the future.</p><p>How does climate change impact wildfires? Since the 1970’s, there has been a global increase in wildfires. With more than 400ppm of CO2 in the atmosphere today, we have seen melting ice caps, increased storm intensity, and higher frequency of natural disasters. Seasons have shifted due to the greenhouse effect, which has caused dry soil in summer and fall. Over the last few decades, the wildfire season has grown by 78 days and is expected to keep expanding.</p><p>What does this mean for the future of wildfires in California? Residents are already starting to see the results of climate change as the last three years were marked as the worst wildfire seasons recorded. For the future, scientists are predicting that all California residents are at risk of wildfires year around. Along with frequent storms, an increase in population is putting residents at higher risk as homeowners are building homes in fire-prone areas.</p><p>What should California residents do to prepare for the future? It is recommended that California residents develop a plan to reduce their exposure to fire by ridding their property of dead material, use fire-resistant building materials in their home, and using flame resistant vegetation on their property.</p><p>If a fire strikes, make sure to have a financial plan to recover and rebuild from fire damage. Check out how Chestnut can protect your home and family for as low as $5 a month. <a href="http://getchestnut.com/">getchestnut.com</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=f5544f162a8e" width="1" height="1" alt=""><hr><p><a href="https://medium.com/chestnut-insurance/the-future-of-wildfires-f5544f162a8e">The Future of Wildfires</a> was originally published in <a href="https://medium.com/chestnut-insurance">Chestnut</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Saddleridge and Sandalwood Wildfires Cause Governor Newsom to Declare a State of Emergency.]]></title>
            <link>https://medium.com/chestnut-insurance/saddleridge-and-sandalwood-wildfires-cause-governor-newsom-to-declare-a-state-of-emergency-7aba5a1072bd?source=rss-3d9e2e9fd419------2</link>
            <guid isPermaLink="false">https://medium.com/p/7aba5a1072bd</guid>
            <category><![CDATA[california]]></category>
            <category><![CDATA[environment]]></category>
            <category><![CDATA[insurance]]></category>
            <category><![CDATA[wildfires]]></category>
            <dc:creator><![CDATA[Anil Celik]]></dc:creator>
            <pubDate>Wed, 30 Oct 2019 17:11:51 GMT</pubDate>
            <atom:updated>2019-10-30T17:11:51.738Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*SK5Y4SasdpZyvUIBJCQryw.jpeg" /></figure><p>(October 18th, 2019) — This past week, California wildfires devastated communities and displaced thousands of families from their homes. The Saddleridge fire broke out on Thursday, October 10th at 9pm. The fire destroyed 19 structures while causing significant damage to 88 homes. Over the course of the week, first responders worked tirelessly to put out fires, but high winds led the fires to burn 8,391 acres and forced 100,000 people to evacuate their homes. Sylmar, Granada Hills and Porter Ranch were the affected areas and the cause of the fire is still under investigation. Inspectors suspect that the blaze began beneath a high-voltage SC Edison transmission tower in Sylmar.</p><p>During the same week, the Sandalwood fire burned 1,011 acres and lasted 7 days, damaging 16 structures and destroying 74. With the Sandalwood fire contained, families are returning back to what they call home to assess the damage.</p><p>Governor Gavin Newsome issued a state of emergency proclamation for Los Angeles and Riverside Counties. This will enable the wildfire victims to obtain assistance from state and federal sources, however, residents are still faced with a heavy financial burden. Rebuilding after a disaster has proved to be difficult with higher insurance premiums and the cost of rebuilding their homes. Many families have had issues receiving their insurance money, while some families are expected to wait months or even years to receive their insurance money. For those who have received their money, the rebuilding time has been stretched as a limited amount of laborers work tirelessly to rebuild entire communities. Families are faced with a lack of resources, money and time to get back on their feet from wildfires.</p><p>Communities are also voicing their concern over authorities promises to make alerts accessible. The residents of Saddleridge are concerned that existing emergency alert systems aren’t properly functioning, meanwhile PG&amp;E had to shut off power during the fires, making it difficult for residents to know when to evacuate.</p><p>As authorities work to help communities, thousands of families are still displaced after the Saddleridge and Sandalwood fires. No one will ever be prepared to face the damage done by wildfires, but there are some things you can do to protect you and your family for the future.</p><ol><li>Analyze your wildfire risk with our <a href="http://analyze.getchestnut.com/">free wildfire analysis tool</a></li><li>Create a wildfire evacuation plan</li><li>Make sure you have encrypted, digital copies of important documents.</li><li>If you have pets, be sure to bring food, water, blankets and any medication.</li><li>Check on loved ones in the affected area.</li><li>Map out shelters in your area.</li><li>Have a financial plan for after a wildfire strikes. <a href="http://getchestnut.com/">getchestnut.com</a></li></ol><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=7aba5a1072bd" width="1" height="1" alt=""><hr><p><a href="https://medium.com/chestnut-insurance/saddleridge-and-sandalwood-wildfires-cause-governor-newsom-to-declare-a-state-of-emergency-7aba5a1072bd">Saddleridge and Sandalwood Wildfires Cause Governor Newsom to Declare a State of Emergency.</a> was originally published in <a href="https://medium.com/chestnut-insurance">Chestnut</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Predicting High Risk Clients Using Machine Learning]]></title>
            <link>https://medium.com/@acelik/predicting-high-risk-clients-using-machine-learning-85f34c9f66bb?source=rss-3d9e2e9fd419------2</link>
            <guid isPermaLink="false">https://medium.com/p/85f34c9f66bb</guid>
            <category><![CDATA[insurtech]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[insurance]]></category>
            <category><![CDATA[machine-learning]]></category>
            <dc:creator><![CDATA[Anil Celik]]></dc:creator>
            <pubDate>Tue, 27 Feb 2018 16:56:22 GMT</pubDate>
            <atom:updated>2018-02-27T16:56:22.417Z</atom:updated>
            <content:encoded><![CDATA[<h4><em>This post was originally published in UrbanStat’s </em><a href="http://www.urbanstat.com/predicting-high-risk-clients-using-machine-learning/"><em>company website</em></a><em>.</em></h4><p>A couple of years ago when <a href="http://www.urbanstat.com/">UrbanStat</a> first started operating, our solutions focused on simulations, deterministic/probabilistic risk analysis, and data visualization. Leveraging a large list of external cat/non-cat data sets and having access to historical datasets from multiple insurance companies enabled us to focus more on machine learning based predictive analytics and create even more value for our carrier partners.</p><p>Our focus was on finding a way to accurately predict high-risk clients, at the time of underwriting, based on their likelihood of claims and damages within a year. Defining “high-risk” is very much dependent on the insurer’s risk appetite as the events we are trying to predict are quite rare and can materially affect the underwriters’ performance. This makes this problem both challenging and exciting as we’ve returned clear risk selection improvements with our partners.</p><p><strong>The graph below visually communicates our goal and value-add in the risk selection process</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*niZHbv2oPZxRBVCfV0vhJg.jpeg" /></figure><p>All insurance carriers have this hypothetical curve (green zone) that represents the ideal clients they want to win. They are OK with paying claims, as long as they are in the green zone (risk appetite). However, carriers end up with this triangle instead (green + blue + red) due to common risk identifying complexities and the lack of proper tools (technology, skillset, team, process etc.) for this problem. Although it’s possible to identify many of the clients that are in the blue zone, It’s almost impossible to identify the ones in the red zone (or as we call internally the god-zone).</p><p><strong>Every client is different so are their risks</strong></p><p>We believe one of the most obvious reasons for missing “right” clients or winning “wrong” clients in the selection process is that carriers are still not granular enough in their segmentation. The problem with this type of segmentation is that many times clients are considered to be similar (if not identical) to everyone else in, let’s say, a given neighborhood so we need to come up with new approaches to enable personalized underwriting.</p><p>Automating a personalized risk selection process for an underwriting team is something we continue to refine and work with our carrier partners on. We continue to focus on identifying high-risk clients that are in the blue section in the graph above so insurance carriers can identify, reconsider or reject clients that pose a greater risk threshold than their internal mandates — as this small percentage of clients are actually threatening the wellness of the portfolio as a whole.</p><p><strong>How to choose policies to be rejected?</strong></p><p>At first glance, one may think it’s an easy question. You should reject policies that are more likely to have higher (how high is defined by the risk appetite) damages than their premiums, which are causing losses to companies. However, when this issue comes to ML, it’s not that easy.</p><p>Let’s assume there is an insurance company X and its main goal of ML usage is maximizing its profit. Theoretically, maximized profit can be reached only by eliminating all policies having higher damages than premiums. So, labeling such policies as class 1 and the rest as class 0 can be a good start for ML. However, since there is a huge class imbalance in this industry (<a href="http://www.urbanstat.com/handling-imbalanced-datasets/">more information</a>) and challenging nature of claim prediction, it can be difficult to have satisfying prediction accuracies. Moreover, false predictions can cause more loss than the profit increase caused by true predictions.</p><p>So, carrier X needs to consider other objectives in which machine learning algorithms will be able to provide ‘optimal’ profit for them. Here are a few objectives decision-makers need to consider:</p><ul><li>Loss ratio</li><li>Actual profit</li><li>Revenues</li><li>Losses</li><li>Market share</li></ul><p>It is quite possible to have great results on loss ratio while losing actual profits, revenues, and market share, or if you just focus on market share, you could actually increase your loss ratio and reduce your actual profits.</p><p>At <a href="http://www.urbanstat.com/">UrbanStat</a>, our objective is to maintain or improve actual profits while reducing the loss ratios with minimal effects on market share. This means by rejecting ~ 3% of your targeted high-risk submissions (minimum loss of market share) you gain an opportunity to improve your profit and your loss ratios at the same time. This approach will enable you to grow in the right direction with increased profits immediately and eventually improve your other financial results and provide better prices and terms to your own clients.</p><p><strong>Approaches are not objectives</strong></p><p>Although your objective stays the same, your approaches can vary. In our case, the most common approach would be regression and classification.</p><p>You can try to predict damage, profit, loss ratio, or a combination of these metrics using regression. After obtaining such predictions, you can easily come up with strategies based on your clients’ “profitability”.</p><p>Another approach would be classifying (identifying) clients whether they belong to a certain group of clients –again by analyzing their damages, profits, loss ratios, number of claims, or a combination of these features. This approach may sound like existing rule-based segmentation algorithms, however, it’s drastically different in terms of its development.</p><p>Depending on the carrier’s historical data, the performance of these different approaches is varying. Our experience shows that classification algorithms work better for many insurance carriers, although one should test both approaches to determine what works better for them.</p><p><strong>So what UrbanStat’s doing differently?</strong></p><p>On average, an insurance carrier provides to us around 100 columns from their clients. We increase the number of columns (features) to somewhere between 600–2000. Gathering and creating external datasets that would explain the damages (e.g. flood hazard maps, hurricane maps, terrorism database etc.) is the first step. We combine this information with carrier’s historic losses. Half of our time is usually spent on engineering additional features than what the carrier provided, where those new features really empower the machine learning applications for success. From our experience, we found that on average 74 of the top 100 features used for prediction are features that we have engineered ourselves. Removing those features makes the models unprofitable immediately.</p><p><strong>Results</strong></p><p>After learning and working on historic datasets of more than 10 insurance carriers, we can confidently say loss ratio improvements of 17% is achievable which some results actually even much higher.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=85f34c9f66bb" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Colors, Methods and Mistakes in Spatial Data Visualization]]></title>
            <link>https://medium.com/@acelik/colors-methods-and-mistakes-in-spatial-data-visualization-60d02e7f09fd?source=rss-3d9e2e9fd419------2</link>
            <guid isPermaLink="false">https://medium.com/p/60d02e7f09fd</guid>
            <category><![CDATA[data-science]]></category>
            <category><![CDATA[data-visualization]]></category>
            <category><![CDATA[gis]]></category>
            <dc:creator><![CDATA[Anil Celik]]></dc:creator>
            <pubDate>Mon, 24 Aug 2015 20:49:24 GMT</pubDate>
            <atom:updated>2015-08-25T19:55:01.233Z</atom:updated>
            <content:encoded><![CDATA[<p>Around ten days ago, I have made a presentation in an event organized by Data Science Istanbul group in Istanbul, Turkey. Seminar focused on data visualization and my presentation covered the theme of spatial data visualization.</p><p>I wanted to share my presentation with everyone, since we weren’t able to host everyone.</p><p>I’ve covered following topics:</p><ul><li>Color Psychology</li><li>Power of Contrast &amp; Boundaries</li><li>Importance of Classification</li><li>Scale: Decision &amp; Visualization</li><li>Interpolation vs Aggregation</li><li>Volume != Area</li><li>Heatmaps != Hot-Spots</li><li>Where “0” Starts?</li><li>%5 of Your Audience</li><li>Worst Mistake</li></ul><p>I hope you enjoy it. Please let me know if you want to add to and/or fix my slides.</p><h3>Color Psychology</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*OG3SeELWTSyJKhtfzNeIfQ.jpeg" /></figure><p>There are many resources online about Color Psychology. Every color has multiple meanings &amp; feelings. Therefore, you should be careful when you use a color for your data visualization. People expect to see rivers in blue, parks in green and danger in red.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*B8WOAmzHdIQWYp_fXqFTZA.jpeg" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*766Skv3i7nCpqnaYTWWJ0g.jpeg" /></figure><h3>Power of Contrast &amp; Boundaries</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*99wc1xLwfKtduwWXbb6Xmw.jpeg" /></figure><p>What you see is usually defined by what you see around it. Your choice of colors should be as contrast as possible and you need to seperate your data points with right visualizations. Some good &amp; poor examples are provided by ESRI. As you see slide above, some maps are harder and some are easier to understand. What makes a visualization is easier to understand is usually the good use of contrast colors &amp; boundaries.</p><h3>Importance of Classification</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*1BJXZw6s1l9Gd68p-b3PhQ.jpeg" /></figure><p>Slide above shows 3 different maps that are visualized from same data using same number of classes. However, all three of them tells a different story.</p><p>First map is classed with Geometric Interval algorithm, that optimizes both the number of data points in a class and variance within the class.</p><p>Second map is classed with Equal Intervals algorithm, that basically divides the data with equal intervals.</p><p>Third map is classed with Natural Breaks algorithm and it keeps the variance within the class minimum while increasing the variance between classes.</p><p>The mission of analyst is to use the correct classification algorithm to pass the message of data. One mistake, and you are telling a completely different story.</p><h3>Scale: Decision &amp; Visualization</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*f9Yiw6dPtR1zi1EXeRmv3g.jpeg" /></figure><p>A mistake that I come across very frequently and did it myself multiple times in the past.</p><p>Usually due to lack of necessary datasets, people tend to use boundary based demographics to explain feasibility of businesses. Map above shows the highly populated neighbourhoods in red and sparsely populated neighbourhoods in green.</p><p>Imagine that you are starting a local business and try to optimize accessible population within 10 minutes walking distance. If you use larger boundaries for your demographic analysis, you are likely to make a big mistake. Because “A” region in yellow neighbourhood actually has more population than “B” region in red zone. I’ve seen banks making decisions with datasets that even have worse resolutions. Big mistake!</p><h3>Interpolation vs Aggregation</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*642nx6vFEhymyab1RzLgyA.jpeg" /></figure><p>First thing we need to know is that interpolation is predictive while aggregation is descriptive and they cannot be used for one another.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*hrSdXqdVeOIAV1C8dnMhlg.jpeg" /></figure><p>Map above shows your customers on a map. Each point represents a customer and you know how much revenue you are making from each of them. You want to see regions that you sell most.</p><p>You can visualize your dots with colormaps to create a more understandable map:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*zcdz57dklaLGpTiiiGGdHg.jpeg" /></figure><p>It’s still really hard to see where you are selling most. However visualization makes you believe that right side of the region has more revenue.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*nwynMeLQiFHTc7Yi2q-xjQ.jpeg" /></figure><p>By gridding the whole region into equal sized areas, we can aggregate our points to improve our visualization.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*BDdUmrywoqODJlg2gWC4sA.jpeg" /></figure><p>Now It’s a lot better to see where we sell most. After this step there is a big pitfall almost every cartographer fall for. We tend to create heatmaps to show visually more pleasing maps.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*ashPcWPnJpe_6E64X8TPxQ.jpeg" /></figure><p>Heatmaps are by default result of different interpolation algorithms and they try to predict areas that we have no value for.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*ASTFnuEiJ2oyNObZXdCS_Q.jpeg" /></figure><p>As seen in slide above, we interpolated our dataset to create visually more pleasing map, however we also broke the data and started to show places that normally have no revenue as generating revenue.</p><h3>Volume != Area</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*4YFUOzVq4cz0FFFMGEDtlA.jpeg" /></figure><p>This is a common mistake that we do everyday. When you look at this map, you feel like company ABC has more market share than the company XYZ. This is due to the landsize of the states. Administrative boundaries are usually meaningless for these kind of datasets.</p><h3>Heatmaps != Hot-Spots</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*MCYAifZTJxoe6st8hUPUMg.jpeg" /></figure><p>It’s often a mistake to use heatmaps. Because everytime you want to use a heatmap, you are probably trying to answer a question that can be answered by another algorithm.</p><p>Heatmaps are always subjective, end-results are highly effected by parameters you set and ignores the statistical significance. On the other hand, hot-spots are objective and uses algorithms to find statistically significant regions where your high values occurs.</p><p>It’s safe to say that in heatmaps, your outliers have greater effect on your end-result as seen on map above.</p><h3>Where “0” Starts?</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*xSn3sZvT0C07QCMFBdBikQ.jpeg" /></figure><p>When you look at this map, you feel like almost half of the area is actually a sea. However due to the color ramp and classification parameters that I used, we see the areas that are not water as water. Correct visualization should be as followed:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*3avNB5Mme0ZhpFc8vBaVVA.jpeg" /></figure><p>It is really important to set the right color for your values that are turning from negative to positive, bad to good, low to high etc. If you pick your color ramp and classification parameters wrong, then you are transmitting a completely different message.</p><h3>%5 of Your Audience</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*ZkdGffdquyeMHbymZHgJPA.jpeg" /></figure><p>Almost %5 of your audience is color blind, most of them are men and green to red color-ramps are nightmare for them. Most of color-blinds see the big map in yellow to dark green tones as you can see in the same slide. Color blinds see those three maps almost identical. There are also other kinds of color blindness which are really rare. If you want to check your visuals for color-blind friendliness, there are great tools online!</p><p>Yes, I know that I made this mistake earlier :(</p><h3>Worst Mistake</h3><p>That’s a mistake that I’ve deliberately done so far.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*lLO4kd1vshhR5hbwmOHspw.jpeg" /></figure><p>Legends are boring but necessary. We often lack them on our visualizations, because somehow we believe that our visuals are perfect and will be understood by everyone in a single look. They are important and if you forget to add legends than you don’t have any excuse for misunderstandings in your visuals.</p><p>These are the common mistakes that I came across and done myself in the past, please share your comments if you have more.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=60d02e7f09fd" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Tehlike Haritaları Yazı Serisi: Sel]]></title>
            <link>https://medium.com/@acelik/tehlike-haritalar%C4%B1-yaz%C4%B1-serisi-sel-cd4a9b09f007?source=rss-3d9e2e9fd419------2</link>
            <guid isPermaLink="false">https://medium.com/p/cd4a9b09f007</guid>
            <category><![CDATA[flood-hazard]]></category>
            <category><![CDATA[sel-riski]]></category>
            <category><![CDATA[insurance]]></category>
            <dc:creator><![CDATA[Anil Celik]]></dc:creator>
            <pubDate>Sat, 18 Apr 2015 07:58:52 GMT</pubDate>
            <atom:updated>2015-04-18T08:48:02.853Z</atom:updated>
            <content:encoded><![CDATA[<p>Sigorta şirketleri ile yaptığımız çalışmalar sırasında Türkiye’de tehlike haritalarının ulaşılması en güç veriler olduğunu farkettik. Bir çok uluslararası şirketin global tehlike haritaları olsa da yerel seviyede analizler yapıldığında bu tehlike haritalarının Türkiye gerçeklerini çok da iyi modelleyemediğini gördük.</p><p>Türkiye’de özellikle sel alanında resmi ve halka açık bir şekilde paylaşılan bir tehlike haritasının olmaması, bu haritaların hidrolik modelleme yöntemleri ile oluşturulmasına olanak sağlayacak debi ve yağış istatistiklerinin ya hiç olmaması ya da kamuyla paylaşılmaması bu alanda yapılacak çalışmalarda farklı bir yöntem arayışına girmemize neden oldu.</p><p>Yaptığımız literatür araştırmaları sonrasında kategorik bir sel tehlike haritası oluşturmak için çalışmalara başladık.</p><p>Mart ayının başında tamamladığımız modelleme üzerinde farklı sigorta şirketlerinden elde ettiğimiz hasar verileri ve global sel tehlike haritaları ile karşılaştırma analizleri yaptık.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*q8-djwTbJxZg_t5BOetBqw.jpeg" /><figcaption>Mekansal İşler Sel Tehlikesi vs Global Flood Hazard Map</figcaption></figure><p>Yukarıdaki haritada gördüğünüz lacivert ve petrol yeşili olan alanlar Mekansal İşler tarafından yapılan modelleme sonucunda çok yüksek ve yüksek sel tehlikesi olan alanları, beyaz ile boyalı olan bölgeler ise uluslararası bir şirket tarafından belirlenen sel tehlikesi olan alanları işaret ediyor.</p><p>İki veriyi farklı sigorta şirketlerinin geçmişe dönük hasar verileri ile test ettiğimizde özellikle Ayamama deresinin kuzeyinde kalan alanlarda ve doğusunda kalan ve Esenler’den başlayıp Bakırköy’e kadar devam eden ve yer yer kurutulan Dereboyu’nda yoğunlaştığını belirledik.</p><p>Adetsel olarak dağılıma baktığınızda tüm bölgelerde irili ufaklı sel hasarları oluşsa da, tutar olarak baktığımızda yüksek riskli olan alanlarda oluşan sel hasarlarının diğer alanlarda oluşan sel hasarlarından ortalamada <strong>8 kat</strong> daha fazla olduğunu gördük.</p><p>Oluşturduğumuz sel tehlike haritası ile sigorta sektörü için geliştirdiğimiz <a href="http://www.mekansalisler.com/sigorta-kumul-risk-takibi-urbanstats/">mekansal karar destek sistemini</a> sigorta şirketleri için yalnızca daha iyi karar vermelerine olanak sağlayacak bir araç olmanın ötesinde teknik karlılıklarını önemli bir şekilde arttırmalarına olanak sağlayacak operasyonel bir araç haline getiriyoruz.</p><p>Tehlike Haritaları olarak kurguladığımız seride bir sonraki konumuz heyelan olacak.</p><p>Bizden daha sık haber almak için <a href="https://www.linkedin.com/company/mekansal-i%CC%87%C5%9Fler">Linkedin sayfamızı</a> takip edebilirsiniz.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=cd4a9b09f007" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Adres — Koordinat Dönüşümü Neden Zor?]]></title>
            <link>https://medium.com/@acelik/adres-koordinat-donusumu-neden-zor-8b7101030bcf?source=rss-3d9e2e9fd419------2</link>
            <guid isPermaLink="false">https://medium.com/p/8b7101030bcf</guid>
            <dc:creator><![CDATA[Anil Celik]]></dc:creator>
            <pubDate>Tue, 16 Dec 2014 20:39:58 GMT</pubDate>
            <atom:updated>2015-03-11T10:23:54.042Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/821/1*koF2G1QZyoxYaAuKOb5pcA.jpeg" /><figcaption>Here Street Maps</figcaption></figure><h2>Adres — Koordinat Dönüşümü Neden Zor?</h2><p>Türkiye gibi gelişmekte olan ülkelerde adreslerin kayıt altına alınması sırasında takip edilen standart bir yöntem yoktur. İşin içine yazım hataları, algısal hatalar, kurumsal veri toplama kaynaklı hatalar girince aynı adres belki onlarca farklı şekilde verilebilir.</p><h3>Şekilsel, Fonetik ve Yazım Hataları</h3><p>Çok basit bir örnek vermek gerekirse;</p><p><strong>Mecidiye Mah. Fıstıklı Köşk Sok. No:25 / 8 Beşiktaş, İstanbul — Türkiye</strong></p><p>adresini aynı binada oturan aileler şu şekillerde verebilir:</p><ul><li>Fıstıklı Köşk. Sok. Numara 25–5 Ortaköy, İstanbul — Türkiye</li><li>Fıstıklıköşk Sk. No:25/4 Ortköy, İstanbul — Türkiye</li><li>Mecidiye Mahallesi Fıstıklı Köşk Sok. 25/3 Ortaköy, İstanbul — Türkiye</li><li>FIstıklıköşk Sokak Mecidye Mah. 25/2 Beşiktas, Istanbul — Türkiye</li></ul><p>Yukarıdaki adreslerde bulunan tüm yazım hataları bilerek yapılmıştır. Herşeye rağmen bu adresler mevcut adres kalitesi dikkate alındığında oldukça “düzgün” sayılabilecek adreslerdir.</p><h3>Algısal Hatalar</h3><p>Diğer yandan kişilerin yaşadıkları mahalleyi ya da ilçeyi yanlış bildikleri durumlar vardır. En sık yapılan hatalardan bir tane örnek vermek gerekirse:</p><p><strong>Ecza Sokak No: 6 Levent, İstanbul — Türkiye</strong></p><p>Levent mahalle olarak bakıldığında <strong>Beşiktaş</strong>’a bağlıdır. Semt olarak bakıldığında <strong>3 ilçenin kesişimindeki bir çok mahallenin</strong> “Levent” olabileceği öngörülebilir ancak semtlerin resmi olarak kabul edilen sınırları yoktur. Ecza sokak ise idari olarak <strong>Esentepe Mahallesi</strong> içinde olup, Esentepe mahallesi ise <strong>Şişli</strong>’ye bağlıdır.</p><h3>Kurumsal Hatalar</h3><p>Bazı kurumlar adresleri sokak, mahalle, ilçe, il gibi kırılımlarda alırlar. Ancak bu bile problemi çözmeye yetmez. Zira farklı kişiler farklı şekillerde veri girerler. Adreslerimizde coğrafi bir hata olmadığını, yazım kurallarına tamamen uyulduğunu varsayalım. Veri toplarken belirlenen kolonlar ise aşağıdaki gibi:</p><p><strong>İç Kapı No; Bina No; Sokak; Mahalle; İlçe; İl</strong></p><p>5; 6; Ecza; Esentepe; Şişli; İstanbul</p><p>Peki kurumunuzda adres alan onlarca insan varsa ve doğru kontrol mekanizmaları ve yönlendirmeler olmadığı için adresleriniz şu şekilde alınmaya başladıysa:</p><ul><li>5; 6; Ecza Sk.; Esentepe Mh.; Şişli; İstanbul</li><li>4; 10; Kore Şehitleri; Esentepe; Şişli; İstanbul</li><li>5; 6; Yıldız Posta Caddesi; Esentepe Mahallesi; Şişli; İstanbul</li></ul><p>** Bazı firmalar adresleri hiyerarşik bir yapıda almaya çalıştığı için kullanıcılardan adres almaları imkansız hale gelmektedir. Kendisini A mahallesinde zanneden ancak C mahallesinde olan bir kullanıcı ilgili ilçe altında A mahallesini göremeyince adres girememektedir.</p><h3>Türkiye’ye Has Hatalar</h3><p>Adreslerinizin bir kısmında konut projeleri ve organize sanayi bölgeleri varsa durum iyice karışmaya başlayabilir:</p><p><strong>Ağaoğlu My City A1 Blok, Ümraniye, İstanbul</strong></p><p>aslında resmi ve ticari veritabanlarında</p><p><strong>Çakmak Mahallesi, Tavukçuyolu Caddesi, 24 C, Ümraniye İstanbul</strong></p><p>olarak yer almaktadır.</p><p>Diğer yandan Türkiye’de sürekli değişen mahalle, ilçe sınırları ve isimlerini de es geçmemek gerekiyor. 2012 yılında ondan fazla kentin büyükşehir statüsüne kavuşması ile bir çok adres ilçe değiştirdi. Çok daha popüler bir örnek olan Ataşehir Barbaros Mahallesinin Ümraniye’ye geçmesi durumu var. Bir mahallenin toptan ilçe değiştirmesi görece takip edilebilir bir durum ancak mahallenin belli kısımlarının ilçe ya da mahalle değiştirmesi durumu işi oldukça karışıklaştırıyor.</p><h3>Gerçek Hayat</h3><p>Gerçek hayatta ise bu hataların her birinin aynı anda yapıldığını düşünün ve çok sayıda müşterisi olan bir kurumun adres veritabanının nasıl gözükebileceğini hayal edin. Veri yapısal olarak alınsa bile adres-koordinat dönüşümü basit veritabanı sorguları ile yapılamaz çünkü her bir kolonda aslında o kolonda olmaması gereken kısaltmalar ve hatta bazen başka kolona ait veriler olabilir. Bu durumda adres-koordinat dönüşümünün en önemli aşaması veri ayrıştırma / temizleme çalışması olmaktadır. Bu işi yapan ulusal ve uluslarası çok sayıda şirket bulunmaktadır. Ancak, kısaltmaların (mah, sk. vb.) olmadığı ve olduğu karmaşık veri setlerinde bu dönüşümü yapabilmek için farklı bir yaklaşım gerektirmektedir.</p><p>Aşağıda kısaltmaların olmadığı bir veri üzerinden İstanbul’un en merkezi yerlerinden birinde bulunan bir adresi farklı şirketlerin altyapılarında test ettik:</p><p>Aşağıda araması yapılan adres aşağıdaki yapıda alınmıştır:</p><p><strong>sokak; semt; ilçe; bina no; il</strong></p><p>fıstıklı köşk; ortaköy; beşiktaş, 25; istanbul</p><p>Bu adrese karşılık gelen resmi adres ise aşağıdaki gibidir:</p><p><strong>Mecidiye Mahallesi, Fıstıklı Köşk Sokak, No:25, Beşiktaş, İstanbul</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/765/1*umLKSRMsNJ8VYGhUb5ecdQ.jpeg" /><figcaption>16/12/2014 Adres Ayrıştırma-Koordinat Dönüşümü Testi</figcaption></figure><p>Bu şekilde alınan bir adresi uluslararası ve ulusal diğer şirketlerin çevrimiçi bir şekilde erişilen harita sayfalarında sorguladığımızda <a href="http://www.mekansalisler.com/">Mekansal İşler</a> Geocoding servisi dışında yalnızca <a href="http://www.here.com/">Here Maps</a> ‘in bulduğu iki adet sonucun ilkinde doğru koordinatları bulduğunu gördük. Google’ın yalnızca Ortaköy’ü algılayıp Ortaköy’ün koordinatını verdiğini, Bing ve Yandex’in hiç bir şekilde sonuç dönmediğini farkettik.</p><p>Tüm bu problemlerin bir arada bulunduğu veritabanları ile yaptığımız testlerde (gerçek müşterilere ait gerçek adreslerle), <strong>minimum %67</strong>, <strong>maksimum %97</strong> ve <strong>ortalamada ise %85</strong> başarı oranı ile çalışan bir <a href="http://www.mekansalisler.com/geocoding">algoritmamız</a> olduğunu bilmek yaptığımız çalışmanın oldukça yüksek bir başarı oranına sahip olduğunu gösteriyor. Ancak yol haritamızda bu başarı oranını ortalamada <strong>%5</strong> kadar daha arttıracağını düşündüğümüz iyileştirme çalışmaları yer alıyor.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=8b7101030bcf" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[İdari Sınıra Bağlı İstatistiksel Çalışmalar ve Sorunları]]></title>
            <link>https://medium.com/@acelik/i-dari-s-n-ra-ba-l-i-statistiksel-cal-smalar-ve-sorunlar-61014d88fea9?source=rss-3d9e2e9fd419------2</link>
            <guid isPermaLink="false">https://medium.com/p/61014d88fea9</guid>
            <dc:creator><![CDATA[Anil Celik]]></dc:creator>
            <pubDate>Sat, 13 Dec 2014 19:56:31 GMT</pubDate>
            <atom:updated>2014-12-13T21:05:14.079Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/954/1*6gtOYxw8cf0Hl-SSympquA.jpeg" /><figcaption>Ekonomik coğrafya idari sınır dinlemez.</figcaption></figure><h3>İdari Sınıra (Mahalle, İlçe vb.)Bağlı İstatistiksel Çalışmalar ve Sorunları</h3><p>Daha önce <a href="https://medium.com/turkish/mekansal-veri-ve-gosterimi-uzerine-c73e446761e7">“Mekansal Veri ve Gösterimi Üzerine”</a> isimli yazımda da bahsettiğim bir konuyu ve çeşitli sektörlerdeki etkisini biraz daha detaylı incelemekte fayda olduğunu düşünüyorum.</p><p>Sosyo-Ekonomik verilerin mahalle ya da ilçe sınırları dikkate alınarak analiz edilmesi bir çok araştırma alanında önemli sorunları da beraberinde getiriyor.</p><p>Bunun en basit örneğini Levent Mahallesi’nde bulunan ve satılık konutların metrekare fiyatları üzerinde yapılan bir analizle görebiliriz. Rastlantısal olarak seçilen yaklaşık 1000 konuta ait basit istatistikleri gösterir tabloyu aşağıda bulabilirsiniz:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/617/1*RJxM10F3GE7r_wfdrKqtVA.jpeg" /></figure><p>Aynı mahalle içerisinde bulunan konutların minimum değeri 2600 TL / Metrekare iken maksimum değer ise 34162 TL / Metrekare olarak görünüyor. Ortalama üzerinden oluşturulan Index ise 15221 TL / Metrekare gibi bir değeri gösteriyor. Index hazırlayan veri sağlayıcıları, danışmanlık firmaları ya da kendi verilerini kullanan finansal şirketler bazen mahalle ölçeğine bile inmeyen istatistiklerle karar veriyorlar. Meşhur “<strong>Gayrimenkul sektöründe balon var mı?</strong>” sorusunu soran yaklaşımın problemi aslında bu bakış açısı. “<strong>Gayrimenkulde nerede balon var?</strong>” sorusunu sormaya başladığımız anda belki de mekansal istatistik (spatial statistics) kavramını öğrenmeye başlayacağız.</p><p><strong>Mahalle ve İlçe Bazlı Verileriniz Ne Kadar Doğru?</strong></p><p>Tüm mahallelerin / ilçelerin mekansal olarak homojen bir yapıda olduğu varsayımını kabul bile etsek, acaba aldığımız mahalle bilgileri ne kadar doğru?</p><p><a href="http://www.mekansalisler.com/">Mekansal İşler</a>’de son dönemde farklı şirketler için yaptığımız bir diğer çalışmada insanların yaşadıkları mahalle ya da ilçe ile ilgili yanlış beyanlarda bulunduklarını farkettik. Farkında olmadan yapılan yanlışlıklar olduğu gibi statü tedirginliği sebebi ile bilerek yapılan “yanlışlıklar” da olduğuna dair bir görüşümüz var.</p><p>Örneğin; Levent’in çevresinde bulunan 4–5 mahallede bulunan kişilerin sıklıkla Levent olarak adres verdiğini gördük. Burada ortaya çıkan temel problem, Esentepe’nin Şişli’de, Emniyetevler’in Kağıthane’de Levent’in ise Beşiktaş’ta olması. Şişli ilçe sınırlarının 2 ayrı parçada yer alması konusuna girmiyorum bile. Bu problemler sebebi ile bir çok “geocoding” çözümü düşük başarı oranları ile adres-koordinat dönüşümü yapabildiği için Mekansal İşler bünyesinde geocoding çalışmasına daha farklı bir <a href="http://www.mekansalisler.com/geocoding">yaklaşım</a> getirdik.</p><p><strong>Nüfus Bazlı Yer Seçim Kararları</strong></p><p>“Gut feeling” dediğimiz Almanlar’da hiç görülmeyen bir hastalık bizim iş hayatımızda virüs gibi yayılmış durumda. Elbette bu verilerle tek başına karar verilemediği için, karar vericiler kendi kişisel tecrübelerine dayanarak karar vermeye başlıyorlar. Umuyorum bu sebeple önemli potansiyellerin kaçtığını ya da ciddi zararlar edildiğinin farkındayızdır.</p><p>Aşağıdaki haritada Kadıköy ilçesindeki mahallelerin toplam nüfuslarına göre renklendirildiğini göreceksiniz. Yeşilden kırmızıya gittikçe nüfus artıyor.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/707/1*p8voB40jLHNCw8LfeQKpyg.jpeg" /></figure><p>Aynı büyüklüğe sahip iki dairenin altında kalan nüfusun bir tanesinin diğerinden <strong>%20 daha fazla</strong> olduğunu ve bu dairenin A dairesi olduğunu belirtsem? Dahası yaya hareketliliği de dikkate alındığında A dairesinin bulunduğu bölgeden muhtemelen çok çok daha fazla nüfus geçeceğini belirtsem?</p><p>Diğer yandan nüfus verilerinin yalnızca mahalle ölçeğinde sağlanıyor olması önemli bir problem olsa da hiç bir zaman aşılamaz bir problem olmadı.</p><p>Soru, görüş ve önerileriniz için Twitter ya da web sayfamız üzerinden bana ulaşabilirsiniz.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=61014d88fea9" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>