Forecasting public transit passenger demand: With neural networks using APC data

Abstract

The implementation of Intelligent Transportation Systems (ITS) as a part of smart mobility is crucial for solving the current problems of the transportation industry. The setting up and maintenance of ITS requires not only the current passenger demand but also the future passenger demand. The future passenger demand can be obtained with time-series forecasting carried out with different techniques. With the advancements in the technological field, modern and more advanced methods of time-series forecasting using deep learning are being preferred over traditional forecasting techniques. However, the research carried out in this regard is quite limited, particularly considering the Indian scenario. Hence this research work focuses on exploring the performance of deep learning forecasting techniques considering the aspects mentioned previously. Here, the forecasting of passenger demand was done with Long Short-Term Memory (LSTM) using the three months Automatic Passenger Counter (APC) data of the Hubballi-Dharwad Bus Rapid Transit System (HDBRTS) as part of a case study. Then the forecasting of passenger demand was also done with Seasonal Autoregressive Integrated Moving Average (SARIMA), and the comparison of the forecasting accuracy of both methods was made using Mean Absolute Error (MAE) and Root Mean Squared Error (RMSE). Furthermore, to validate the results, novel approach has been adopted for the process, by following some more time-series resampled with different time intervals. Study shows that LSTMs will be used satisfactorily in the traffic conditions of developing counties, for forecasting passenger demand using APC data. Study also provides detailed guiding methodologies of advanced methods of passenger forecasting along with conventional ones.