Vladimir Cherkassky, Professor, Computer Science and Engineering
The goal is to develop accurate real-time prediction of highway traffic density during congestion periods, using current and time-lagged observations of volume and occupancy from detector stations. Accurate traffic prediction is critical for effective control of on-ramp traffic (ramp metering). Software for real-time traffic prediction will be developed based on advanced Statistical and Neural Network methods for nonlinear time series prediction. Prediction performance of several state-of-the-art methods, such as MARS, Projection Pursuit Regression, back propagation neural networks and Constrained Topological Mapping will be evaluated using real traffic data. The results of this study will evaluate predictive modeling capability of these advanced methods for real-time traffic prediction.