A SOFT COMPUTING APPROACH TO TRIP GENERATION ESTIMATION IN LAGOS METROPOLIS, NIGERIA

Abstract

Trip generation is an indispensable component of the four-stage transportation planning process because the subsequent three stages are predicated on its results.  Linear regression has been widely adopted to predict trips due to its simplicity and its outperformance of more sophisticated count models and in some cases, soft computing models. The efficacy of regression for estimating trip generation alongside Artificial Neural Networks (ANN) and Fuzzy Expert System (FES) was examined. The performance of each model was evaluated using metrics such as Mean Absolute Error (MAE), Mean Squared Error (MSE), Coefficient of Determination (R2) and the capability of predicting average trips. The R2 for Regression, ANN and FES were all 0.71. The MAE for Regression, FES and ANN were 0.56, 0.55 and 0.49 respectively. The MSE for Regression, ANN and FES were 1.15, 1.16 and 1.15 respectively. Finally, FES and ANN resulted in average trips of 4.5 in comparison to actual average trips of 4.51 per household,  while regression produced average trips of 4.51. ANN and FES are not superior alternatives to the linear regression model for trip generation modelling. The performance increments gained from adopting these models are marginal and the extra development and computational effort required to apply such sophisticated approaches may not be justified