Evaluating vehicle driver classification: a supervised machine learning approach through comparative analysis

Abstract

Contemporary computing and applications of data science chime in unison when it comes to machine learning. Thus the boundaries of learning using AI is open for expansion, aiding those with limited resource and compact scope. Extensively, Supervised Learning has indigenous uses for predictive mapping. Using multi-class vehicle driver data as apparatus, this paper targets to analyze the classification accuracy, precision, predictability, rigidity and extrapolation of existing supervised learning algorithms in case of subjective mechanical data and objective qualitative data. Driver data is compiled focusing on the classical mechanics of driven vehicles and external affecting factors in every driving instance followed by processing for feature selection to split data for training and testing. Analyzing the variance of learning via different supervised machine learning algorithms, a shortlisted precedence of algorithm preference is prepared through comparative exploration. The exploratory extractions are then analyzed to reach optimal extrapolation of vehicle driver classification within the domain of driving style, adjudged distinctly as aggressive, normal and vague. Achieved insights would actively contribute to solidify drivers' conformity towards traffic laws and situational safe driving, aiming to secure a significant fall in road casualties.