Analysis of Public Transport with Time Series Forecasting

Anže Schwarzmann (2018) Analysis of Public Transport with Time Series Forecasting. MSc thesis.

Preview

PDF Download (6Mb)

Abstract

Public transport is a vital part of each major city, since it enables simple and environmentally-friendly transportation throughout the city. It is expected that the rising number of inhabitants will increase the demand for public transport. When using the city buses, the passengers especially see the unpredictable arrival times as one of the largest flaws. In order to improve the quality of public transport, time series forecasting methods have been chosen to analyse how accurately we can foretell the number of passengers at a bus stop and the bus travel times between two bus stops. We were also interested in finding a possible connection between the forecasted accuracy and unpredictability of the number of passengers and bus travel times. We had data of bus arrival times to the bus stops and number of passengers at bus stops at different hours in Ljubljana. Different time series forecasting methods have been used in our research. We also analysed the influence of various weather conditions on the accuracy of our prediction. It has been indicated that when predicting the bus driving period or especially the number of passengers at bus stops, good knowledge of the weather conditions in the near future, e. g. in next hours, can significantly influence the improvement of prediction accurate. This was presented using ARIMAX method that uses explanatory variables together with other classical methods for time series forecasting. The following were chosen as representatives of other classical methods: (1) AR – Autoregressive model, (2) ARIMA – Autoregressive integrated moving average model, and (3) VAR – Vector auto-regressive model. ARIMAX method is the only of the listed prediction methods that assumes that we know the future values of some of the variables that are, however, not the subject of our prediction. The experiments have shown that in cases when we use weather data for explanatory variables, the ARIMAX method leads to the better and more consistent predictions. As expected, the predicted accuracies were better in summer months and at weekends, when the number of passengers is smaller and when bus driving periods are shorter. The same is true for the bus stops that are not located in the city center. We have proved that the number of passengers at a bus stop does not have a particular influence on improving the accuracy of predicting bus travel time between two bus stops. The experimental results indicate that the increased traffic does not necessarily mean more unpredictable driving periods.

Actions (login required)