Abstract: Aiming at the low accuracy of the motorized traffic mode identification in the residents traffic mode identification, this paper proposes an improved LightGBM algorithm combined with a traffic mode classification method of mobile terminal. This method filters the data set, selects the time domain and frequency domain features of three kinds of sensor data: triaxial accelerometer, gyroscope and magnetometer as the pattern recognition features, and uses the Filter correlation measure CFS algorithm to sort the scores according to the features, and selects the optimal feature set. The recognition process adopts the K-lightGBM recognition algorithm based on the residents travel rules and the first-order hidden Markov chain. At the same time, some machine learning algorithms are used for comparison experiments. The experimental results show that this method not only can identify multiple modes of traffic, but also has a high average accuracy of residents traffic pattern recognition, reaching 94%.

Key words: sensor, CFS, first-order hidden Markov chain, LightGBM algorithm； traffic pattern identification