TTC Transit Service Reliability Analysis

This project analyzes 23,701 real TTC delay records to uncover what actually predicts transit service disruptions. It includes a custom station-name normalization function (achieving a 99.6% match rate across inconsistently recorded station names), a comparison of Random Forest and Gradient Boosting models for delay prediction (~0.62–0.65 ROC-AUC), and a 5-feature experiment set including a leakage-safe rolling delay frequency feature. Findings are presented through a full EDA notebook and a 3-page interactive Power BI dashboard, aimed at supporting data-informed decisions on transit reliability.