TRAVEL MODE CLASSIFICATION BASED ON CELLULAR DATA

Cellular data could be one of the primary data sources of transportation planning because of the rapid growth in triangulation and location techniques. This study aims to develop a classification algorithm to accurately predict the mode used by the mobile user by using genetic fuzzy logic control (GFLC) model. Four state variables are chosen, including trip length, travel speed, bus trajectory similarity, and rail trajectory similarity. The consequent part is the mode classification among four modes: bus, rail, private vehicle, and non-motorized vehicle. To facilitate the training and validation of the proposed model, this study invites 50 volunteers to join a 30-day travel diary survey. The accuracy rate based on 5-fold cross-validation is between 74% to 86%. For the prediction performance of the proposed model, the highest accuracy ratio is rail, followed by private vehicles and non-motorized modes. The correct prediction rate of bus is the lowest.