Development of Machine Learning Model to Predict Driver's Subjective Evaluation of Tire During Lane Change Operation

Abstract

It is important for tire developers to clarify the judgment basis of subjective evaluation results. In this research, we constructed a machine learning model that uses sensing data related to the driver’s perception and operation as input and subjective evaluation as output. The model combined ESN and random forest to clarify important features and timings from whole time series data of all features. The internal state of the ESN was taken out at an arbitrary timing so that it was input to the random forest, and the subjective evaluation values were predicted. Prediction of the values at lane change showed sufficient accuracy. From the importance of the feature values, it was found that the steering torque in the early stage of lane change, the steering angle and steering torque in the middle stage, and the movement of the head in the late stage. The approach used in this research can clarify important features and timings, so it can be applied to various situations using time-series data.