PRAHARI: Advanced Predictive Safety Platform
Year: 2026
Type: article
Abstract: —Navigation apps often provide routes based on distance and not much else. Many factors can impact how safe a route is for a female walking at night or a student unfamiliar with an area in the city. PRAHARI is a web application that takes this into account. We ask two questions: How safe are the road segments on each route? How much extra distance would we have to travel to avoid the less safe segments? Using the OpenStreetMap database, we create a directed graph containing all roads. Each road segment has 7 different attributes mapped including the class of road, length of road, number of lanes, one-way, illumination tag, distance to nearest police station, and distance to nearest nightlife spot. An XGBoost regressor was fitted to give each segment a score between 1-10. Dijkstra's algorithm provides us with a shortest, safest and balanced route. The navigation service was built using FastAPI with a backend containing node.js/express and a frontend using react/leaflet. This allows us to also track the journey, report incidents and send SOS requests. For the central section of Bengaluru containing 160,077 segments, 32 s was taken to build and score the area. Each segment had a prediction. However, a problem with the original design was found while going from our campus to Majestic. The 'safest' route was 151% longer but only reduced the risk index by 1. Since we were using a cost per segment, routes consisting of fewer, longer segments were favoured. We explore why this is happening. We also looked at an alternative of normalising the distance, and tested this offline using the same data.
Authors: Mr. Abhishek B M, Priya, Aditya Kumar, Ashwin Kumar
Institution: Acharya Institute of Technology
Language: English
Subfield: Automotive Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7225896437
