Transportation mode choice prediction using a new multi-class association rule model

This research focuses on multiple class association rule (MCAR) approach to study travel behaviour. In MCAR approach, how to utilise a top-k rule team set to build an effective classifier remains a challenge for travel behaviour research. Another challenge is that each top-k rule team has only one class label, which may lead to a few observations being predicted incorrectly. To tackle the challenges, this research utilises a CNN model and similarity between observations (SBOs) to obtain multiple class labels for a top-k rule team. Finally, a set of top-k rule teams with multiple class labels is utilised for developing a similarity-based MCAR (S-MCAR) model to predict transportation mode choices. To evaluate the effectiveness of the developed model, a travel dataset and 5 comparison models are utilised. In addition, in the S-MCAR and comparison models, a ten-fold cross-validation method and a grid search method are applied.