Novel optimised deep learning approach for an efficient traffic state prediction based on CAE-ICCDCS-GRU model

Deep neural networks (DNNs) have recently been proven capable of forecasting traffic flow from large datasets. While existing DNN models can outperform shallow models, it is still a work in progress to adjust the hyperparameters of DNN models using an effective optimisation approach. Therefore, an improved crossover-chaos-based dynamic cuckoo search (ICCDCS) algorithm has been developed to improve the traffic state prediction efficiency. In the proposed traffic state prediction model, convolutional auto-encoder (CAE) is employed to achieve efficient noise filtering with feature extraction, and the gated recurrent unit model (GRU) uses ICCDCS to optimise its hyperparameters for forecasting traffic status in the most efficient way possible. To increase the effectiveness of the cuckoo search, ICCDCS combines the Circle map for chaos, crossover operation, and boundary management method. These combined features make it possible to select a better GRU hyperparameter, improving the accuracy of traffic state forecasting. Faster searches and better information interchange are ICCDCS benefits. The efficiency of ICCDCS algorithm has been tested with test functions. Other optimisation techniques are less accurate than ICCDCS. The result of CAE-ICCDCS-GRU reveals drastic improvement in the performance, apparently in terms of minimised error due to its optimised parameters of GRU.