<?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 G.Rushindhra Goud on Medium]]></title>
        <description><![CDATA[Stories by G.Rushindhra Goud on Medium]]></description>
        <link>https://medium.com/@2303a52199?source=rss-3693eb26364c------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/0*QsHD2YBODAcdBts8</url>
            <title>Stories by G.Rushindhra Goud on Medium</title>
            <link>https://medium.com/@2303a52199?source=rss-3693eb26364c------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Sat, 21 Nov 2026 10:04:25 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@2303a52199/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[Can a Map Warn You Before You Reach a Dangerous Road?]]></title>
            <link>https://medium.com/@2303a52199/can-a-map-warn-you-before-you-reach-a-dangerous-road-e85565c37edb?source=rss-3693eb26364c------2</link>
            <guid isPermaLink="false">https://medium.com/p/e85565c37edb</guid>
            <category><![CDATA[data-science]]></category>
            <category><![CDATA[machine-learning-python]]></category>
            <category><![CDATA[road-safety]]></category>
            <dc:creator><![CDATA[G.Rushindhra Goud]]></dc:creator>
            <pubDate>Wed, 18 Nov 2026 17:46:13 GMT</pubDate>
            <atom:updated>2026-11-18T17:46:13.742Z</atom:updated>
            <content:encoded><![CDATA[<p>How our capstone team turned 5,000 accident records from Warangal into live risk alerts with machine learning</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*cD1-l_SlnxQryhbo-ZY69g.png" /></figure><p>SafeRoute maps accident hotspots and rates the risk of a trip. Image by the author.</p><p>In 2023, India recorded 480,583 road accidents that killed 172,890 people, about 20 deaths every hour (<a href="https://morth.nic.in/sites/default/files/Road-Accident-in-India-2023-Publications.pdf">MoRTH, 2023</a>). Over-speeding alone was linked to about two-thirds of those deaths.</p><p>These crashes are not spread evenly. They cluster on black spots: stretches of national highway about 500 metres long with five serious accidents or ten deaths in three years (<a href="https://eparlib.nic.in/bitstream/123456789/1058601/1/AU764.pdf">MoRTH definition</a>). The data to find them exists, but it sits in PDF reports that drivers never open.</p><p>Our final-year capstone project, SafeRoute, set out to close that gap.</p><h3>What we built</h3><p>SafeRoute is a web dashboard that shows where accidents concentrate around Warangal and rates how risky a trip is before you start. You choose a location and switch three conditions on or off: night-time driving, drunk driving and over-speeding. The app adds the live weather, runs a machine-learning model and returns a Low, Moderate or High risk level.</p><p>The answer never arrives as a bare label. Each result shows a confidence score, the probability of each risk level and the factors that pushed the risk up, followed by a colour-coded safety alert.</p><h3>How it works</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*FVKERqgUZfK0pfv5RafRcw.png" /></figure><p>How data moves through SafeRoute, from raw records to the alert on your screen. Image by the author.</p><ol><li><strong>Data.</strong> We started with 5,000 accident records, each with 43 attributes. After cleaning, we engineered 10 new features, such as a 0–4 weather-risk score, for 53 columns in all.</li><li><strong>Black spots.</strong> We grouped nearby accidents into zones and flagged the zones with the most crashes and casualties as black spots on the map.</li><li><strong>Prediction.</strong> An <a href="https://arxiv.org/abs/1603.02754">XGBoost</a> classifier, a model built from many small decision trees, predicts the risk level from the location’s history and the conditions you choose.</li><li><strong>Serving.</strong> A <a href="https://fastapi.tiangolo.com/">FastAPI</a> backend with a SQLite database answers the dashboard and fetches live weather from <a href="https://openweathermap.org/current">OpenWeatherMap</a>. The React dashboard draws the map with <a href="https://leafletjs.com/">Leaflet</a> and runs on Vercel.</li></ol><h3>Results</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*UhA06_gs7mM03PYfV7nB2g.png" /></figure><p>Accuracy and F1 score of the four models we compared. Image by the author.</p><p>Logistic regression gave us a baseline of 79.5% accuracy. Random forest reached 83.9%, and tuned XGBoost did best, at 84.6% accuracy with an F1 score of 0.844, so it became our final model.</p><p>In one test, we checked Subedari at night, in moderate rain, with drunk driving and over-speeding switched on. SafeRoute predicted Moderate risk with 72% confidence and listed five contributing factors, including the spot’s closeness to a known black spot. Because of that history, it also raised a High Risk Alert.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*j1XttYNqlzc-ce_De43_rw.png" /></figure><p>A risk check for Subedari on the live SafeRoute dashboard. Screenshot by the author.</p><h3>What I learned</h3><p><strong>Accuracy is not enough.</strong> Even at 84.6%, the model is wrong about 15 times in 100. Showing the confidence and the reasons lets people judge an answer instead of trusting it blindly.</p><p><strong>Plan for failure.</strong> A weather service can go down or an API key can go missing. SafeRoute falls back to “Clear” weather, so a failed call never blocks a prediction.</p><p><strong>Small details break deployments.</strong> A model saved with one scikit-learn version may not load under another, so we pinned every library version. On free hosting, the first visit after a quiet spell is slow while the server restarts and rebuilds its database.</p><p><strong>Agree on the interface early.</strong> Four of us split the model, backend, dashboard and deployment. Fixing the API’s request and response formats first let us all work in parallel.</p><h3>What’s next</h3><p>SafeRoute runs on one city’s historical data for now. Next, we want to connect it to official police accident records, cover more districts and warn drivers along a whole route, not just at one location.</p><p>Which road in your city would you check first? Tell me in the comments.</p><p><em>SafeRoute was built by G. Sai Teja, G. Rushindhra Goud, G. Vishnu Vardhan and S. Kirthan under the guidance of Dr. Rajnish Kumar, Associate Professor, School of CS&amp;AI, SR University, Warangal.</em></p><h3>References</h3><ol><li>Ministry of Road Transport and Highways, <a href="https://morth.nic.in/sites/default/files/Road-Accident-in-India-2023-Publications.pdf">Road Accidents in India 2023</a>.</li><li>Parliament answer citing the MoRTH black-spot definition, <a href="https://eparlib.nic.in/bitstream/123456789/1058601/1/AU764.pdf">circular dated 28 October 2015</a>.</li><li>T. Chen and C. Guestrin, <a href="https://arxiv.org/abs/1603.02754">XGBoost: A Scalable Tree Boosting System</a>, KDD 2016.</li><li><a href="https://fastapi.tiangolo.com/">FastAPI documentation</a>.</li><li><a href="https://leafletjs.com/">Leaflet</a>, an open-source JavaScript library for interactive maps.</li><li><a href="https://openweathermap.org/current">OpenWeatherMap Current Weather API</a>.</li></ol><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=e85565c37edb" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>