Commuter train service quality prediction and user attribute ranking using neural network and fuzzy approach

Abstract

Railways have always been playing a vital role in economic growth and development of Bangladesh by hauling both goods and passengers. Bangladesh Railways has been facing many alarming challenges from alternative modes of transport during the past decades. In this regard they have taken an initiative to reduce the congestion of the city road and ensure a safe journey by introducing commuter train service in different routes in the country. This research is aimed to asses users perception about this commuter train service. The service quality (SQ) of commuter train has been carried out by developing two models named (i) Adaptive Neuro Fuzzy Inference System (ANFIS) (ii) Probabilistic Neural Network(PNN) in this research. For this purpose a questionnaire survey has been carried out among 802 respondents whom are travelling in commuter train as their mode of transport. This questionnaire was comprises in four sections (i) personal and socioeconomic information (age, gender, occupation etc) of the respondents (ii) 24 questions about the service quality (SQ) of commuter train (iii) rating the overall SQ of commuter train (iv) selecting at least 10 questions or attributes as the most significant attributes. Data collection has been conducted at Dhaka, Rajshahi, Narayangonj. The respondents were asked to number every question by 1 to 5 where “5” corresponds to excellent quality and “1” corresponds to very poor quality. About 61% of the total respondents rate the overall SQ of commuter train as satisfactory or good and 20% thought that the SQ of commuter train is poor. From the data of questionnaire survey 12 attributes have been selected for model development. In prediction of Service Quality (SQ) ANFIS has given 61.50% accuracy in training period which is about 395 of total 641 predictions and in testing period the accuracy is 47.80% which is about 77 out of 161 predictions which matches with the actual value. In PNN the accuracy is 67.40% in training period which is 433 out of 641 predictions and 44.10% in testing period which is 72 out of 161 predictions which matches with actual Service Quality (SQ). Finally, a stepwise approach was followed for ranking the commuter train SQ attributes influencing its overall SQ and the results were compared with that of the public opinions. Attribute ranking was conducted by using Root Mean Square Error (RMSE) and correlation coefficient R values and comparing these values with the public opinion about the attributes which are most significant. From the results, it is found that 'Bogie condition', 'Cleanliness', 'behavior of staff', 'toilet facility' are the most significant attributes. This indicates that some necessary measures should be taken immediately to recover the effects of these factors to improve the service quality of commuter train. This research came to a result that ANFIS is most suitable for commuter train service quality analysis than PNN model.