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

Abstract

In megacities of developing countries, the need for mobility is increasing in synchronization with the growth of the cities themselves. Conversely, mobility and accessibility are decreasing hastily and it is most severe in case of public transport (PT) users. Currently, in developing countries, the real problem is not the high use of automobiles, but the poor PT service quality (SQ). It is seen that, the services provided by transportation operators of developing countries may not necessarily satisfy passengers’ expectations. Like-wise, a developing country like Bangladesh has PT vehicles that are frequently poorly maintained and often overloaded. Particularly, the only requirement is to fulfill the need for mobility with sufficient capacity. Whereas, the quality is constrained by the government’s limitation. In that case, the real contribution of paratransit becomes significant. Among the different available PT modes, paratransit plays a vital role, especially where there is insufficient mass transit system. Paratransit is recognized in Dhaka city as a special transportation service with higher flexibility and availability in selected routes operated by private companies as well as individuals. This research aimed to assess users’ perception of this PT mode. Moreover, several empirical models were developed to predict its SQ. Through these data driven models, the variables influencing the paratransit SQ were determined, which could lead to improve the overall paratransit SQ of the developing countries. At first, this study examined fifteen strategic locations for fifteen different paratransit service routes in Dhaka city to collect the required data to assess the overall SQ of this mode and to formulate empirical models. In this context, a stated preference (SP) survey was conducted among the paratransit users in each survey location. For the data collection, the designed SP questionnaire comprised of two sections, where (1) The first section was aimed to get personal and socioeconomic information (age, gender, occupation) of commuters and the reason for using paratransit mode; and (2) the second section was focused on twenty three (23) questions regarding paratransit SQ (twenty-two SQ attributes and a question about overall paratransit SQ) to know the actual conditions of this mode in Dhaka City. All the questions about the paratransit SQ were in a close-ended format with relevant multiple choices those were chosen by the users. It was found that major portion (42%) of the respondents rated the overall quality of paratransit service ‘satisfactory’ while 30% users’ thought that existing condition is good and 22% opined that it is in poor condition. Based on users’ perception and the stated ratings (22 paratransit SQ attributes), it was found that majority of the user opined that the following factors are the advantages of using paratransit service: (i) Cleanliness of the vehicle; (ii) Speed of the vehicle; (iii) Availability of vehicle; (iv) Travel time (Holidays); (v) Integration with supporting modes; (vi) Security of goods; (vii) Travel cost and (viii) Service feature. However, there were some following factors identified by the user, which are the main limitations of paratransit: (i) Meager seat comfort level of paratransit; (ii) Substandard fitness of the vehicle; (iii) Dissatisfactory noise level of the service; (iv) Insufficient lighting facilities; (v) Inconvenient ticketing system (fare collection) to the users; (vi) Unskilled paratransit drivers; (vii) Risky entry-exit system; (viii) Congested sitting arrangements for passengers; (ix) Inadequate movement flexibility in the vehicle; (x) High travel time during office day; (xi) Not enough security of the passenger during off-peak period; (xii) Poor riding safety; (xiii) Ordinary performance of long route movement; (xiv) Low graded movement flexibility of vehicles in any road. With the inadequate resources, developing countries like Bangladesh will find it difficult to invest in improving all of the significant attributes’ quality as were found from this study at once. This investigation provides guidance for a stepwise development which will start with the most important attribute. Based on the users’ stated preferences (on a scale of 1 to 5), two Artificial Intelligence (AI) models namely Probabilistic Neural Network (PNN) and Adaptive Neuro-Fuzzy inference System (ANFIS) were developed using a dataset extracted from 2008 paratransit users. These models can predict the paratransit SQ based on twenty two (22) attributes. A comparison on the prediction capability between PNN and ANFIS was also presented. The comparison results showed that PNN outperformed ANFIS. Particularly, the coefficient of correlation (R) values of PNN and ANFIS prediction were 0.702 and 0.442, respectively. Whereas, the Root Mean Square Error (RMSE) values for those models were 0.745 and 0.929, respectively. The study was further extended to include ranking of the SQ attributes according to their significance. This was necessary to identify the key attributes affecting the paratransit SQ. Out of 22 SQ attributes, ‘Ticketing system (Fare Collection)’, ‘Quality of Driver’, and ‘Security of passengers’ were found to be the top three attributes having the most influences on the users’ decision making process. All these findings can aid city transportation officials and service providers in improving the most important paratransit attributes, thereby increasing its ridership.