Optimized Deep Neural Network Based Intelligent Decision Support System for Traffic State Prediction

Importance of efficient short term traffic state prediction has been increased for accurate traffic planning in the domain of an Intelligent Transportation System. Modeling variety of traffic patterns and unanticipated traffic flow changes with time dependencies are the primary problems in traffic prediction. Existing approaches suffer to capture non-linearity of traffic flow complex features efficiently. Therefore, an intelligent decision support system for traffic state prediction has been proposed to boost the efficiency of the traffic state prediction model. Spatio-temporal based optimized Gated Recurrent Unit (GRU) has been developed to implement an intelligent decision support system for traffic state classification. Initially spatial features are learnt using the Convolutional Neural Network (CNN) model. Traffic state is predicted using GRU where the hyper parameters of GRU degrade the performance of traffic state prediction. Therefore, GRU is integrated with Grasshopper Optimization Algorithm (GOA) for the regulation of the hyper parameters in GRU. The CNN-GRU-GOA model was evaluated with CNN-LSTM, LSTM and Stacked auto encoder. The CNN-GRU-GOA achieves 96.8% of accuracy in PeMs dataset and 96.7% of accuracy in china traffic dataset which reveals that performance of traffic state prediction has been enhanced drastically by CNN-GRU-GOA with less computational cost.