Publication Date

Spring 2026

Degree Type

Master's Project

Degree Name

Master of Science in Computer Science (MSCS)

Department

Computer Science

First Advisor

Maryam Khazaei

Second Advisor

Melody Moh

Third Advisor

Amith Kamath Belman

Keywords

Machine Learning, Deep Learning, Traffic Collision, Injury Prediction

Abstract

Studying traffic collisions and the conditions in which they occur offers insight into how we can better design vehicles as well as the roads they drive on. The city of San Jose has been recording collisions occurring in the city since 1977. This project focuses on collisions from 2010 and after to account for changes in road conditions. Using machine learning classification techniques, it may be possible to predict the severity of the injuries sustained by those involved in a collision based on historical data of similar collisions in the city. This project implements a variety of traditional machine learning models like Logistic Regression and XGBoost as well as several CNN and other deep learning architectures. Due to severe injuries being fortunately rare in collisions, the injury classes are severely imbalanced. This is reflected in the experimentation results where the accuracy is high but the recall/precision for the positive class is low. Attempts to account for this through synthetic sampling and class weighting showed some improvement. The results of this project ultimately show that traffic collisions have many factors outside of what’s included in this dataset that influences the severity of injury.

Download

Available for download on Wednesday, May 26, 2027

Share

COinS