Traffic Prediction Using Auxiliary Information Based on Mlbsae-A Hybrid Deep Learning Framework

Traffic prediction is one of the challenging tasks and has many applications. The effective predictive data analytic algorithm gives the smart city concept wings and is essential for controlling chaotic traffic conditions. Due to the non-linear and chaotic features of large traffic data, Deep Learning techniques can play major role in increasing the effectiveness of predictive analytics in Traffic Modeling. Taking into account the benefits and drawbacks of various prediction models, this study proposes a hybrid deep learning framework based on BiLSTM and stacked autoencoders. We called this model MLBSAE (Multi-Layer BiLSM and Stacked Auto-Encoders). Complex chaotic features from traffic data are extracted using stacked autoencoders (SAE). Output of the feature extraction produced by stacked autoencoder becomes input sequence to the multilayer BiLSTM. Experiments are conducted on traffic data from California that includes auxiliary information like weather, accident and incident data apart from the traffic data collected through sensors. Recent prediction models like SARIMA, XGBoost and Random Forest models are used to compare the effectiveness of the suggested hybrid model.To evaluate the effectiveness of the models, statistical error measures such as accuracy, precision; F-Score and recall of both micro and macro are considered. The results of the simulation show that MLBSAE generates more precise projections compared to the state of art models.