Analyzing the performance of different machine learning methods in determining the transportation mode using trajectory data

With the widespread advent of the smart phones equipping with Global Positioning System (GPS), a huge volume of users’ trajectory data was generated. To facilitate urban management and present appropriate services to users, studying these data was raised as a widespread research filed and has been developing since then. In this research, the transportation mode of users’ trajectories was identified based on their raw GPS data. These data are often associated with errors, it was attempted to minimize these errors by applying a comprehensive pre-processing procedure in this research. Accordingly, various features were extracted to identify transportation modes including walk, bike, train, bus, and driving. In this regard, four classification methods including decision tree, multilayer perceptron neural network, Naïve Bayes, and support vector machine were used to build a predictive model. In order to improve the performance of the implementation methods, the percentage of points of each trajectory in the distance of one standard deviation from the total speed average of transportation modes has been used as a new feature. The above-mentioned four models were implemented with different regularization parameters and their values were set to the optimal values by applying a comprehensive grid search. Then, Kappa, overall accuracy, and RMSE indices were employed to evaluate different methods. The results of this study show that the multilayer perceptron neural network with overall accuracy of 0.88 have the best results compared to other models.