Back to the Past: Predicting Expected Ride Demand from Previous Dropoffs and Beyond

In the realm of transportation network companies (TNCs), the utilization of demand forecasting models and algorithms has revolutionized essential services, such as guiding idle vehicles, efficient carpooling, rate fluctuations, and price surging. In this work, we investigate the effect of varying learning models and input features on the prediction of future demand. Specifically, we extract simple yet key spatial and temporal features of pickups and dropoffs from NYC taxi data, including day of the week, time within the day, and location, and run them through multiple learning models with varying attributes and complexities, such as regression and Long Short-Term Memory (LSTM) models. Our results highlight the capabilities of harnessing accessible and simple features to enhance ride-demand prediction models, paving the way for more inclusive and transparent advancements in urban transportation services.