A COMPARATIVE STUDY OF CATBOOST AND ARTIFICIAL NEURAL NETWORKS IN ENHANCING TRIP GENERATION MODELLING FOR ILORIN CITY

Abstract

Trip generation plays a crucial role in transportation planning, and the choice of an appropriate model is essential for predicting future travel patterns. This study focuses on comparing the suitability and performance of CatBoost and ANN for trip generation (production and attraction) modelling of Ilorin City. By incorporating Ilorin household and trip characteristics, population data, and maps, this study evaluates the performance of the models. The two models demonstrated high accuracy and performance. In terms of trip production, the CatBoost model displayed exceptional accuracy, attaining an R-squared value of 0.99999992016446, accompanied by an impressively low mean squared error (MSE) of 3.93870930136429e-05. In contrast, the neural network exhibited a slightly lower accuracy of 0.999873850524181, with an error value of 0.0581313408911228. Similarly, for trip attraction, the CatBoost model showcased remarkable accuracy and precision, achieving an accuracy score of 0.9999999999999994 and an extremely low error value of 2.26762031965784e-13. The neural network model demonstrated an accuracy of 0.99999999990335 and a negligible error value of 0.000000041994. These findings underscore the strong predictive capabilities of both models for trip production and attraction, with the CatBoost model notably excelling in achieving nearly flawless accuracy and minimal error values across both aspects in Ilorin. Further research can explore the application of other advanced machine-learning techniques and combine their strengths to enhance the accuracy and robustness of trip-generation models.