Identification Of Driver Types Using Drive Lab Data

Abstract

With a computer-aided driving simulator, the data of the driver’s every move will be recorded. Because of the economic and risk-free feature, driving simulator is becoming popular in vehicle design, road construction and human factors researches. In this regard, we have three hypotheses: 1. Drivers’ performance in a driving simulator can reflect the real driving situations; 2. Different driving behaviour can reflect different personality; 3. Drivers can be classified based on the data collected by a driving simulator. To test these hypotheses, the data from two psychology experiments was applied. The driving bahaviours of different drivers when passing a same hazard were compared and analyzed. An unsupervised K- means was used to give preliminary labels and the artificial neural network was applied to verify the labels. Finally, a significant gender effect was found. Training the neuron network with only female data then save the model to test only male data showed a low predicting accuracy. Repeating the experiments with only female data gave the best accuracies.