An Intelligent Machine Learning Framework for Accurate Uber Ride Fare Prediction Using Gradient Boosting Regression
Authors-Dr. M. A. Azeem, Yatakarla Sandhyarani
Keyword-Uber Fare Prediction, Machine Learning, Gradient Boosting Regressor, Random Forest, Linear Regression, Intelligent Transportation Systems, Ride-Hailing Services, Regression Analysis, Predictive Analytics, Flask Web Application.
The rapid expansion of ride-hailing platforms has transformed urban transportation by offering convenient, flexible, and on-demand mobility services. As the number of daily ride requests continues to increase, accurately estimating ride fares has become essential for improving opera-tional efficiency, enhancing customer satisfaction, and supporting intelligent transportation man-agement. Reliable fare prediction enables passengers to estimate travel costs before booking while assisting service providers in optimizing pricing strategies and resource allocation. This research presents a machine learning-based framework for predicting Uber ride fares using historical trip information and environmental attributes. The proposed framework utilizes public-ly available Uber trip records collected from the NYC Open Data repository, containing infor-mation such as travel distance, journey duration, time of day, and fare values. Before model development, the dataset undergoes comprehensive preprocessing involving missing value han-dling, feature engineering, normalization, and data partitioning to improve data quality and model performance. Three regression algorithms, namely Linear Regression, Random Forest Regres-sor, and Gradient Boosting Regressor (GBR), are implemented and evaluated under identical experimental conditions. Hyperparameter optimization is performed for the Gradient Boosting model by adjusting the number of estimators, learning rate, and tree depth to maximize predictive accuracy while preventing overfitting. Model performance is evaluated using the Coefficient of Determination (R²) and Mean Squared Error (MSE). Experimental analysis demonstrates that the Gradient Boosting Regressor achieves the highest predictive performance, producing a train-ing accuracy of 99.99% and a testing accuracy of 91.79%, outperforming both Random Forest and Linear Regression models. Furthermore, feature analysis identifies travel distance and time of day as the most influential variables affecting fare estimation. The optimized prediction model is integrated into a Flask-based web application that enables users to obtain real-time fare esti-mates through an interactive interface. The proposed framework provides a scalable, accurate, and practical solution for intelligent ride fare prediction and supports the development of efficient transportation management systems.
Doi-[https://doi.org/10.5281/zenodo.21735912]