Prediction and Optimization of Ride-Sharing Routes Using Recurrent Neural Network Models

The study looks into the possible uses of machine learning (ML) technologies, such as Recurrent Neural Networks (RNNs), in optimizing urban mobility, ride-sharing systems. Advanced deep learning (DL) models can solve the emerging issues in metropolitan areas, namely traffic congestion, environmental pollution, and the ineffective public transport systems. SimpleRNN, GRU, and LSTM are three distinct RNN architectures that are compared in this study to forecast the locations of trip pickup and drop-off locations. According to the evaluation measures, LSTM outperforms GRU (9.50%) and SimpleRNN (3.85%) in terms of prediction accuracy, with a Coefficient of Determination score of 10.28%. In addition, LSTM surpasses the other models, displaying the shortest Mean Haversine Distance (6.527 km) and the lowest Mean Absolute Error (0.0212). Moreover, the study integrates these predictive models with real-time geospatial data, using the OSRM API to optimize ride-sharing routes. The research emphasizes how ML, and DL techniques in particular, may be used to address urban mobility issues by enhancing transit efficiency and lowering traffic. The findings give insights for creating more intelligent, flexible city transport networks as well as a data-driven strategy for improving ride-sharing services.