An Effective Prediction Modelling of Traffic Accidents based on Categorical Boosting with Grid Search

Recent days, the alarming rise in road traffic accidents has devastating consequences globally, causing a significant surge in injuries and fatalities. This trend is impaired by projected doubling of vehicles worldwide by 2040 which is placed by unprecedented strain on transportation infrastructure. Traditional approaches for modelling traffic accidents had significant challenges which include difficulty in integrating diverse data sources, complexity and difficulty in understanding model decisions. Therefore, this research proposes Categorical Boosting (CatBoost) integrated with Grid Search for predictive modelling of traffic accidents. Initially, the data is taken from Road Traffic Accident (RTA) dataset which is taken from department of police, Addis Ababa Sub city. Then, this data is preprocessed by using Random Projection Paradigm (RPP) and Improved Linear Discriminant Analysis (ILDA) which effectively reduced dataset dimensions, simplified complex traffic data. After that, the model is selected by using CatBoost which effectively processes the categorical data and ensured robust modeling, the hyperparameters are tuned by using grid search optimization. Finally, the model is evaluated by using k-fold cross validation provided the robust evaluation and handled imbalanced data. The proposed CatBoost method established better results including accuracy (0.95), precision (0.78), recall (0.65), RMSE (0.02) when compared with existing Naïve Bayes.