GPS-based travel survey has been proven effective in transportation mode recognition. Existing studies seldom pay attention to the effects of traffic conditions on transportation mode identification. Besides, due to the similar speed characteristics of buses and cars, the recognition accuracy of the two modes needs to be further improved. Therefore, this paper proposes a procedure optimization method based on Support Vector Machine (SVM) to identity transportation modes of the GPS data collected under different traffic conditions. First, the new frequency domain features generated from Short-time Fourier Transform (STFT) are merged into a pool of features with the traditionally used time domain features. Second, Genetic Algorithm (GA) is used for optimizing the penalty parameter and the nuclear parameter of SVM jointly. Finally, SVM is applied to identify the transportation modes and mode transfer time under different traffic conditions. Results show that the recognition accuracy of the proposed framework outperforms traditional approaches. Traffic states have a great impact on the accuracy of recognizing buses and cars. In the free flow state of the roads, the accuracy of recognizing the two motorized modes can reach up to 91%, and the mode transfer time errors are within 30 s. When the roads are severely congested, the motorized modes are easier to be confused with the non-motorized modes. The maximum error of mode transfer time is within 13 minutes, which may be still informative compared with traditional manual questionnaire surveys based on subjective recall.

CICTP