Safer Routing

Overview

Shortest-path directions can take pedestrians through poorly-lit alleys, high-crime blocks, or stretches with nowhere to turn for help. This project is a routing service that finds the safest path across a city — walking and/or transit — trading a small amount of extra distance for a meaningfully safer route.

The Challenge

• “Safety” isn’t a single signal — it has to be assembled from open businesses (safe havens), historical crime (danger zones), current traffic conditions, & the type of road segment (tunnel, bridge, back alley, avenue)
• Optimizing purely for safety can produce absurd detours; the route still needs to be practical
• Multi-modal routing — walking &/or subway — multiplies the size of the search space
• Validating that routes are measurably safer across an entire city, not just on a few hand-picked examples
• Risk analysis weather to take the shortest route & minimize exposure of a high risk vs deviating slightly for longer time through lower risk paths.

Approach

• Modeled the city’s road & transit network as a graph, with edge weights derived from safe-haven proximity, crime history, live traffic, & road type
• Empirically tuned the weighting so safety gains aren’t dominated by unrealistic detours
• Ran a modified A* search over the graph to find the safest route between any two points, walking &/or by subway
• Exposed the routing engine as a REST API service
• Validated at scale with Monte Carlo simulations over randomly sampled origin-destination pairs within 5km across the city

Results

• Routes scored 2x safer (custom metric) than the shortest path, on average
• Average detour overhead of less than 100m versus the shortest route
• Avoided bridges, tunnels alleys where possible. Preferred busy streets with open shops.