RADIAL BASIS FUNCTION NEURAL NETWORK AND REGRESSION MODELS APPLICATION TO TRAVEL DEMAND ANALYSIS IN AKURE METROPOLIS, NIGERIA

Abstract

With the understanding that the formulation of transportation plans for an urban area are dependent on appropriate and efficient travel demand analysis, this research was aimed at carrying out a comparative assessment of the efficiency of Radial Basis Function Neural Network (RBFNN) and Multiple Linear Regression Models (MLR) in carrying out travel demand analysis for Akure metropolis. The study area was divided into Low, Medium and High residential land use density zones while data for the study were collected from these zones through an intensive household questionnaire interview survey between October 2017 and January 2018. The study specifically focused on the trip generation and mode choice modelling stages of the travel demand analysis process. Statistical Package for Social Sciences version 22 (SPSS 22), was used in carrying out descriptive statistical analysis, correlation analysis as well as model calibration. The RBFNN and MLR were used in developing home-based trip generation models and their performance and accuracy where compared using Mean Absolute Percentage Error (MAPE) and Coefficient of Determination (R2). Mode choice modelling was carried out using the Multinomial Logistic Regression Model (MNL) so as to be able to determine the proportion of home-based trips embarked on using the different modes in the study area. Pearson Correlation analysis showed number of household members, (NHM), number of employed household members, (NEHM), number of students in household (NSH), number of Household members with age greater than 12years (NHM12) and number of Driver’s license holders in the household, (NDLH) as the major factors influencing home-based trips in the study area. Results of the trip generation modelling process for each of the zones showed that the RBFNN with higher R2 values and lower MAPE performed better than the MLR in predicting the number of home-based trips generated in the three zones. In the model of trip generation developed with aggregate of the data from the three zones, the RBFNN predicted 11,147 home-based person trips as against observed value of 11,125 from the survey whereas the MLR predicted 11088 person trips. In the aggregate model, the RBFNN with R2 value of 0.947 and MAPE of 0.391 was also found to be more accurate than the MLR with R2 value of 0.589 and MAPE of 0.875 and was therefore adopted for further analysis. The MNL model used in Mode choice modelling gave a prediction of the mode proportions used for the generated trips as 18%, 46%, 15%, 1%, 20% for Private car, Taxi, Motorcycle, Bus and Walking modes respectively. Expansion of the daily home-based person trips generated for the 98,044 household population of Akure yielded 511,776 daily home-based person trips while a ten year projection of these trips to the year 2028 yielded a total of 713,241 home-based person trips, thereby indicating that there will be about 30% increase in the number of home-based trips generated in the study area by the year 2028. The high R2 values and low MAPE served in validating the accuracy of the models. The study recommends large scale infrastructural development to cater for the 30% increase envisaged in the number of home-based trips in the study area while also advocating the application of artificial intelligence methods to other field of transportation since this study has been able to uphold the capability of artificial neural networks to produce good results in travel demand forecasting areas.