Stacked LSTM for Short-Term Traffic Flow Prediction using Multivariate Time Series Dataset

Short-term traffic flow prediction has paramount importance in intelligent transportation systems for proactive traffic management. In this paper, a short-term traffic flow prediction technique has been proposed based on a Long Short-Term Memory (LSTM) model, which analyzes the multivariate traffic flow data set. To predict the traffic flow of a particular road, it considers present day and historical traffic data of that particular road and also considers historical as well as present day traffic flow data of other dependent neighboring road segments. Stacked LSTM model has been used for accurate prediction of traffic flow for all days, irrespective of weekday or weekend. Simulation has been done on real traffic data, and the proposed technique has been compared with other state-of-the-art techniques to predict the traffic flow. The simulation result shows that the proposed technique forecasts near accurate traffic flow prediction for both weekdays and weekends under both usual and unusual traffic conditions.