In this paper, we describe a non-linear embedding technique called Barnes-Hut Stochastic Neighbor Embedding (BH-SNE) that emphasizes the local similarity structure of high-dimensional data points and applies it to the classification of driving behavior. The mobile terminal collects acceleration sensor data, gyroscope sensor data and magnetometer sensor data during vehicle travel, fuses the three sensor data, performs pre-processing, uses BH-SNE to complete dimensionality reduction processing, and finally inputs dimensionality reduction data into the Radial Basis Function neural network (RBFNN) to identify seven driving behaviors. The experimental results show that the efficiency of BH-SNE is much higher than that of t-Distributed Stochastic Neighbor Embedding (t-SNE), and the visualization effect is better than t-SNE. The overall recognition rate of this classification model is 98.8%, and the classification effect is better than the traditional machine learning algorithm.