Hybrid CNN-LSTM for Traffic Flow Forecasting

Abstract

Listen

Translate article

Accurate traffic flow forecasting is a crucial need for Intelligent Transportation Systems (ITSs). This supports dynamic and proactive traffic control management. The challenging part lies in reducing the forecast error rate. There was limited success in the previous attempts put forth to develop traffic flow forecasting systems. In this work, the MIDAS Site—UK Highways data is used to predict the traffic flow using the hybrid CNN-LSTM model. The data is preprocessed using Z-Score Normalization in order to scale the data below and above the standard deviation. The features from the preprocessed data are learnt using the CNN model and with these extracted features as input, the LSTM network forecasts the traffic flow. The CNN model, LSTM model and hybrid CNN-LSTM model are trained with the preprocessed data. The trained models are tested for estimating the traffic flow. The results convey that the hybrid CNN-LSTM model proposed forecasts the traffic flow with a reduced error rate.KeywordsTraffic flow forecastingConvolutional Neural Network (CNN)Long Short-Term Memory (LSTM)Hybrid CNN-LSTM modelIntelligent transportation system