eduzhai > Applied Sciences > Computer Science >

Multitask Learning Deep Neural Networks to Combine Revealed and Stated Preference Data

  • KanKan
  • (0) Download
  • 20210425
  • Save

... pages left unread,continue reading

Document pages: 25 pages

Abstract: It is an enduring question how to combine revealed preference (RP) and statedpreference (SP) data to analyze travel behavior. This study presents aframework of multitask learning deep neural networks (MTLDNNs) for thisquestion, and demonstrates that MTLDNNs are more generic than the traditionalnested logit (NL) method, due to its capacity of automatic feature learning andsoft constraints. About 1,500 MTLDNN models are designed and applied to thesurvey data that was collected in Singapore and focused on the RP of fourcurrent travel modes and the SP with autonomous vehicles (AV) as the one newtravel mode in addition to those in RP. We found that MTLDNNs consistentlyoutperform six benchmark models and particularly the classical NL models byabout 5 prediction accuracy in both RP and SP datasets. This performanceimprovement can be mainly attributed to the soft constraints specific toMTLDNNs, including its innovative architectural design and regularizationmethods, but not much to the generic capacity of automatic feature learningendowed by a standard feedforward DNN architecture. Besides prediction, MTLDNNsare also interpretable. The empirical results show that AV is mainly thesubstitute of driving and AV alternative-specific variables are more importantthan the socio-economic variables in determining AV adoption. Overall, thisstudy introduces a new MTLDNN framework to combine RP and SP, and demonstratesits theoretical flexibility and empirical power for prediction andinterpretation. Future studies can design new MTLDNN architectures to reflectthe speciality of RP and SP and extend this work to other behavioral analysis.

Please select stars to rate!


0 comments Sign in to leave a comment.

    Data loading, please wait...