AI-Powered Real-Time Dynamic Pricing Decision Tool for Ride-Hailing Platforms

Urban ride-hailing has expanded rapidly over the past decade, but a surprisingly persistent problem remains: passengers have almost no reliable way to compare fares across providers before they book. Demand forecasting is still difficult, pricing is opaque, and the gap between what a trip actually costs and what a user expects to pay can be substantial. To address this, we developed an integrated AI-powered tool that brings fare estimation, demand modelling, and driver availability together in one place, letting users compare real options across providers before making a booking decision. The proposed system combines a Random Forest regression model trained on large-scale historical trip data with a Long Short-Term Memory (LSTM) deep learning model to capture sequential demand patterns over time. Rather than relying on straight-line distance estimates, the system computes routes over actual road networks — a distinction that turns out to matter considerably in practice. A surge pricing layer derives multipliers from the predicted demand-to-supply ratio, and the final output is a structured comparison of ride options across providers and vehicle types — complete with price, arrival time, and the likelihood that a driver will accept the request. The system is deployed through a FastAPI backend, enabling real-time inference. Experimental evaluation using January 2023 NYC Yellow Taxi data demonstrates that the Random Forest model achieves a Mean Absolute Error (MAE) of 1.12 and RMSE of 4.15 for fare prediction, outperforming a Linear Regression baseline. The LSTM model effectively captures temporal demand trends across hourly intervals. Taken together, the results suggest that the technical gap between ‘a collection of ML components’ and ‘a tool passengers can actually use’ is smaller than it appears. Closing that gap — which is what this work attempts — is, we believe, worth considerably more attention from the research community.