A comparative study of machine learning models for taxi-demand prediction using a big data framework

The increase in urban vehicle numbers and ownership extends traffic congestion, with road expansion often escalating the problem, while growing individual car usage further contributes to the issue. To address these challenges, the development of efficient and accurate demand predictors, particularly for taxi-passenger demand, has attracted a significant attention from both industry and academia. This research aims to address the question of how predictive models, evaluated within a big data framework, can be effectively utilized to forecast taxi-passenger demand and enhance urban mobility. The study is structured around two key objectives: first, to evaluate existing predictive models for analyzing traffic data and forecasting taxi-passenger demand using a dataset from the New York City Taxi and Limousine Commission (TLC), and second, to provide practical recommendations for improving demand prediction through a comparative analysis of machine learning models—such as multinomial logistic regression, generalized linear regression, random forest, and decision tree algorithms. Our results demonstrate that the glr outperforms both random forest and decision tree models in predicting taxi-passenger demand, achieving approximately 90% $$\text{R}^{2}$$ R 2 score. The multinomial logistic regression classifier predicts demand levels with over 70% accuracy. The lower performance of random forest and decision tree algorithms is attributed to the highly imbalanced nature of the demand, leading to overfitting and local optimization issues. The application of these predictive models can contribute to optimizing fleet management, reducing passenger waiting times, and improving taxi distribution, potentially alleviating congestion and supporting a better integration with other transport modes.