The research on prediction models for urban family member trip generation

Abstract

This paper validates the hypothesis that head of household's attributes and travel behavior having effects on other family members' trip frequency in the same household. With using Multinomial Logit Model (MNL) and data source obtained from Resident Trip Survey in Nanjing, 2010, it shows that other family members' trip frequency are much more relative to head of household's. Therefore, we propose an idea that investigates the potential possibility of using Multinomial Logit model (MNL) and Support Vector Machine (SVM) to predict other family members' trip frequency with household's information. By comparing the prediction accuracy, it is found that the SVM only slightly performs better on overall average accuracy. Then a modified SVM model with significant variables estimated by MNL model is given. The results report that, modified SVM model not only performs better on overall average accuracy than general SVM model and MNL model, but significantly improves partial average accuracy compared with general SVM model. However, the modified SVM still expectedly has a poor performance on predicting for small proportion samples. In the end, some possible improvements are discussed as well as expectation for further studies.