LSTM Network Integrated with Particle Filter for Predicting the Bus Passenger Traffic

Abstract

The paper reports a combination of the deep learning technique and bayesian filtering to effectively predict the passenger traffic. The architecture of the model integrates the particle filter with the LSTM network. The time series sequential prediction is best achieved using LSTM network while Markovian behaviour is well extracted using Bayesian (Particle Filter) filters. The temporal and spatial features of the traffic data are analyzed. Three relevant temporal variations viz. , morning, noon and post noon patterns are identified after the histogram analysis. These patterns are statistically modelled and the integrated model is used to accurately predict the passenger flow for the next thirty days, facilitating, the bus scheduling for that period. The experimental results proved that the proposed integrated model with coefficient of determination ( ) value of 0.88 is functional in predicting the passenger traffic even when the training data set size is small.

Deep Learning Based LSTM Model for Predicting the Number of Passengers for Public Transport Bus Operators

“…SMAPE calculates accuracy based on error percentage [52] that can be used to evaluate the prediction performance of time series data sets [53] (3). The lower the SMAPE value of a forecast, the higher the accuracy [29].…”

“…A long short-term memory (LSTM) network model was proposed for predicting new passenger flows for public bus transportation with better accuracy results compared to other baseline methods [28]. The Long Short Term Memory (LSTM) model is used to accurately predict passenger flows for the next thirty days [29]. The BiLSTM model is used to predict bus passenger demand based on actual patronage data obtained from the smart card ticket system in Melbourne with accuracy results of more than 90% which can outperform the LSTM model [30].…”

“…Low prediction accuracy results in poor performance of the prediction model which greatly affects the performance of public transportation buses [34]. The accuracy of predicting the number of passengers using deep learning [35] directly influences decisions in the operation and management of public bus transportation [36]. Predictions of the number of bus public transportation passengers are still weak in capturing and measuring passenger distribution [37] with still low accuracy [38].…”