Assessment of the driver's driving style using supervised machine learning

Abstract

Supervised machine learning can be an effective method of predicting a target variable, depending on current values of independent attributes, if historical data containing values of independent attributes and target variables in the past is available and if some machine learning algorithm gives good results on the available data set. In this research, the aim was to show whether the method of supervised machine learning can be successfully applied in assessing the driver's driving style in terms of fuel consumption, if as independent attributes, i.e. parameters that affect the assessment of driving take: engine speed, percentage of accelerator pedal pressure and vehicle acceleration. Training, validation, testing and application of machine learning models were performed in the Weka software tool. The following seven machine learning algorithms were applied to the data sets for model training and testing: LinearRegression, MultilayerPerceptron, IBk (k-nearest neighbors), M5P, Random Forest, Random Tree and REPTree. The best performance was shown by models based on the IBk and Random Forest algorithms. As a final result of this research, predicted scores of the driver's driving style in the interval of one second were obtained. The final score of the driver's driving style was calculated as the arithmetic mean of the predicted driving scores for each second.