A Comparative Analysis on Machine Learning Techniques for Driver Distraction Detection

Abstract

Driving any transport vehicle carries a significant amount of risk. Driving, on the other hand, is a necessity that cannot be ignored or replaced with something safer. The only thing that can be done in this concern is by taking precautionary measures to make driving safer and reducing the injuries and death to almost 0. The distracted driving model was built with the aim of achieving accident-free roads, with no threat and danger of any accident. This model accomplishes this task by detecting the behavior of the driver while driving the vehicle and inform the driver of any distraction which can possibly result in an accident. For this purpose, the machine learning model is trained to categorize the real-time received images of the driver into various classes of distractions. The distracted driver model has been developed on various different models. The accuracy of the predictions made by each of the models is compared to finally achieve the model which can best suit the requirement and provide highly accurate results. After performing a validation check on all the algorithms, KNN model has best accuracy and found out that KNN over LDA had ~99% accuracy which will give near about accurate detection of drivers’ distraction. We also validated it on PCA and PCA over LDA too where PCA gave the accuracy of ~81% PCA over LDA gave ~76%, but LDA gave us the best accuracy, i.e., ~99%. The authors can assist the government in detecting drivers who engage in practices such as drunk driving, rash driving, texting, calling, eating, drinking, and many more by properly implementing this model. The laws can be implemented successfully and easily using this paradigm because no human intervention is necessary. The paper discusses about building a machine learning model for detecting distractions of the driver while driving. The paper comprehensively discusses the steps and methodology required for creating such model. Finally, results were calculated for respective algorithms and based on those selections can be made as per the requirement of the user. It is difficult to choose the best algorithm or the technique that will fit as per the user’s requirement. Therefore, this paper tried to draw a comparison among the various techniques by calculating the values of different parameters and helps the users to select the best technique as per their requirement. Selection of the technique would vary for different users.