Application of Electronic Ticket Machine Data for Analysis and Forecasting of Bus Transport Demand

Abstract

Passenger demand estimation and forecasting is the fundamental process in public transport planning activity. The optimised operational planning of public transport requires the accurate prediction of the passenger flow on a periodical basis. This research tries to develop an intra-city and inter-city passenger demand estimation model using the data-driven time series approach. In this study, the source of time-series data is the Electronic Ticket Machine (ETM) data, which was not previously explored for passenger demand modelling and forecasting by the researchers. The data used for the study are mainly ETM data from Kerala State Road Transport Corporation (Kerala SRTC). During the analysis of the Kerala ETM data, it was found that the length of intra-city ETM data (for Trivandrum city) available for individual routes varies from 1.5 years to 2 years (2011 to 2013), while inter-city data were not sufficient for timeseries modelling. Therefore, inter-city ETM data of Mangalore division Karnataka State Road Transport Corporation (Karnataka SRTC) was collected for a period of 2013 to 2018, which comprises of a total five-year data, to perform inter-city passenger demand modelling. The intra-city ETM data for Mangalore was not collected since the intracity services are mainly operated by private operators. The private operators have not maintained and stored the ETM data. Since the travel demand is affected by the performance of the public transport system and the bus transport agencies operating the services, this study evaluates the performance of the public transport system and the public transport operator. Therefore, in addition to ETM data, this study also uses data provided by various state and central government agencies. The level of service offered by the public transport system in Trivandrum city, the accessibility of public transport and the performance evaluation of Kerala SRTC was performed to evaluate the present condition of the public transport system in Trivandrum and Kerala SRTC. The performance evaluation of Karnataka SRTC was not performed since the data collected from Karnataka was only for modelling inter-city passenger demand. The Level-of-Service (LoS) was determined using the service level benchmarks set by the Ministry of Urban Development, Government of India. This approach can be used to determine the LoS at the city-level and thus provides a measure for identifying the public transport quality of an urbanvi area. The overall LoS is found to be ‘one’. This indicates that the city has good service coverage, i.e., the public transport ply on most of the corridors in the urban area. The facilities are available wide-spread and are available to the people. This study proposed a methodology for the determination of the accessibility index, which is based on the factor that the index should measure the accessibility which comes from proximity to bus stops and land use destinations, and the proportion of the population served. It was observed that the accessibility index in the central area has a higher value and the accessibility values decrease to the periphery of the city. It can be interpreted that the availability of opportunities is higher in the central area, while the number of opportunities is less in other areas. The density of the road network is more towards the city centre, which contributes to higher accessibility values. Also, given that the concentration of population is high in the city centre and the higher number of bus stops, the accessibility index tends to be more in these areas. The performance evaluation of Kerala SRTC was performed using the methodology developed in this study. A weighted Goal Programming (GP) methodology integrated with Analytical Hierarchy Process (AHP) considering the user’s and operator’s perception was used. The operator cost and staff per schedule were the most important variables in regard to the operator, while the safety of travel with respect to user perception. The optimal solution indicates that increasing the accessibility, safety and regularity attract passengers to the public transport that in turn improves the load factor which influences the operators to maximise the fleet utilisation and reduce the cancellation of schedules. The proposed model solution suggests decreasing the staff per bus that will further reduce the staff cost and hence the operating cost. The correlation analysis of ETM data revealed that the relevant temporal pattern in the time-series data (for both intra-city and inter-city) was daily time-series pattern, which was further taken into account during the modelling process. Also, in the data analysis, it was found that 85% of the intra-city data was nonstationary, while 100% of the data for inter-city was nonstationary. The nonstationary data were differenced to make it stationary for further time series analysis. Only 30% of the intra-city data was having seasonal effect while 100% of the inter-city data exhibits seasonality. It is due to the fact that the seasonal variations are not that prevalent in urban travel demand. But intervii city travel is affected by seasonal variations. 95% and 99% of intra-city and inter-city data respectively were non-normal. Therefore, data transformation was required to make these data normal for further analysis. 90% of the intra-city routes were having nonlinearity while all the routes of the inter-city were nonlinear. Hence, nonlinear modelling was required to model it. A methodology to develop an empirical passenger demand estimation model using the data-driven time series approach, employing the capabilities of AutoRegressive Integrated Moving Average (ARIMA) method is developed. The ARIMA model is suitable when dealing with a nonstationary time-series data. But, it cannot be applied to data sets having volatility. When the time-series data has heteroscedasticity, i.e., the variances of error terms are not equal, or some of the error terms are reasonably large, AutoRegressive Conditional Heteroscedasticity (ARCH) models are used for modelling the data sets. When the data have non-standard features like non-normal distribution, and non-linear relationship with lagged values require non-linear AutoRegressive Neural Network (ARNN) modelling methods. The methodology to improve the developed model using the nonlinear ARCH and ARNN methods are determined. It includes a comprehensive methodology for model selection and forecasting, which requires only basic econometric knowledge. The empirical results were compared with the actual value for 30 days and 7 days forecast horizon; and found that the time series model could predict the future observations with an acceptable forecast accuracy. The forecasting performance of developed models are measured using the Mean Absolute Percentage Error (MAPE), and the model goodness of fit is determined using information criteria. The lower value of MAPE represents a good forecast. Lewis (1982) describes MAPE greater than 50% as inaccuracy in forecasting, 20-50% as reasonable forecast, 10-20% as good forecast and less than 10% as highly accurate forecast. In this study, the MAPE values for the 30 days and 7 days forecast were less than 10%. This time-series methodology is useful when there is limited or no information available about the factors affecting the demand. Recommendations have been made to improve the performance of bus operating organisations.