CrashAlert: Predictive Analytics for Road Safety & Emergency Response

CrashAlert is a predictive system designed to analyse and forecast accident risks based on weather, road, and traffic conditions. Built using PySpark and Spark MLlib on over 7 million US records, the system classifies high-risk zones and suggests the nearest hospitals for emergency care — a prototype for potential in-car deployment.

The Problem

Each year, millions of road accidents occur due to a dangerous mix of weather, traffic, and road conditions — often without any real-time warnings or post-accident guidance. Drivers are left unprepared both before a crash and in the critical minutes after. What if cars could anticipate danger and assist in recovery?

The Solution

CrashAlert is a prototype for an intelligent in-car assistant that does exactly that — combining historical accident data with real-time road and weather inputs to both predict dangerous driving conditions and route drivers to nearby hospitals if an accident occurs.

Built on over 7 million U.S. records, the system uses:

• Descriptive analytics to identify patterns behind high-risk scenarios

• Predictive modeling (Random Forest) to flag potential crash moments with 94.2% accuracy

• Geospatial KNN to recommend the top 5 closest hospitals for emergency response

• Visual dashboards (heat-maps, risk maps) to aid deployment in a vehicle interface or driver safety dashboard

While the current implementation is batch-based, the pipeline was architected to scale for real-time streaming and could be deployed in modern connected vehicles or emergency dispatch systems.

Tech Stack

• Data Processing: PySpark (Spark SQL, DataFrames for cleaning and feature engineering)

• Modeling:

  • Random Forest (accident prediction)

  • KNN (nearest hospital recommendation)

• Visualization: Heatmaps and US-wide geospatial mapping for high-risk zones

• Data Sources: U.S. accident, weather, and traffic datasets (7M+ records)

For more details, checkout my Github page: https://github.com/pparashar21/accident-hotspot-analysis