A Methodology for Taxi Demand Prediction Using Stream Learning
David Faial, Flávia Bernardini, Edwin Benito Mitacc Meza, Leandro Miranda, José Viterbo
Abstract
Intelligent transport support systems have had a major impact on people's urban mobility. In large urban centers, transportation services still need ways to optimize vehicle supply in certain areas, according to the demand in each of them. Optimized distribution of on-demand taxi services can be part of an intelligent urban mobility plan, causing direct impacts on urban traffic, improving transport accessibility, improving safety at taxi standpoints by reduced waiting times, reduce transportation fare etc. Many vehicle-mounted sensors currently generate real-time information that is not used for processing and generating information with value. This paper proposes a Taxi Demand Forecasting methodology using stream machine learning algorithms that tackle concept drift detection on taxi data stream. A real data source made available on the New York open platform feeds a stream learning model, constructed using the Massive Online Analysis (MOA) tool - a framework for data stream mining. The stream model shows promising results in forecasting taxi demand, reaching 78% accuracy. Despite using data from a specific city, the methodology and results of this work can contribute to a more proactive demand management in other cities.