Predicting Future Traffic Congestion from Automated Traffic Recorder Readings with an Ensemble of Random Forests

摘要

Predicting future traffic congestion has the potential to decrease travel times by improving GPS navigation and enhancing traffic flows. This paper describes a methodology developed to predict future automated traffic recorder (ATR) readings with current and recent ones from local ATRs. Data was preprocessed by downsampling the simulated ATR signals. Additional training sets were created by resampling the ATR signals at incremental offsets. Regression Random Forests were trained to predict future ATR recordings, and an ensemble of Random Forests created with different preprocessing techniques was formed. This ensemble performed within the top 1% in one track of the 2010 IEEE IGDM Contest: TomTom Traffic Prediction for Intelligent GPS Navigation, improving 43.4% on the baseline algorithm.