Development of drowsiness detection system using machine learning and image processing techniques

Abstract

Since drowsy driving is a major problem in Kenya, which has led to multiple fatal accidents, it has raised discussions that seek for a set of solutions. Several fatalities and injuries resulting from road accidents caused by drowsy driving have been recorded and it has been identified to be a significant problem. Studies that have been conducted in this field have identified interventions in terms of detection and alert systems that can offer solutions to this problem. This study mixes these categories to come up with a machine learning algorithm that records the driver, analyses their condition, processes the information and gives feedback to get the driver back to normal. The data used in training the machine learning model used for this study was extracted from secondary sources, including the internet. The data was then cleaned before use. However, the accuracy of the model determines its application. The model developed for this study generated a weighted average precision of 86% with a recall of 83%, an 84.5% accuracy and an F1-score of 83%. These results are relatively high compared to many models that have been used by previous researchers. This shows how applicable the model is in real situations and how much the other models need to be improved. However, limitations like small size of training data and low processing power of the equipment used should be addressed in future studies for better outcomes.