Detection and Analysis of Drowsiness using Machine Learning

Drowsiness is a major source of fatigue, which can endanger the lives of drivers and other passengers. Teachers are confronted every day with students not paying attention in class because they can be seen falling asleep, yawning and wandering off, which ultimately reduces their attention span and becomes less engaged in class. With workplaces online due to the current COVID situation and employees preferring to work from home during their free time, it's clear they can feel sluggish. Our real-time drowsiness detection system is designed to detect human drowsiness based on body changes, i.e. Open/close eye and mouth status, then alert that person with an on-screen alert. In addition, the sleepy behavior of the person is analyzed using data visualization to understand the sleepy behavior of the person, and the sleepiness report of the person is emailed to the people. The labeled dataset is prepared from real-time facial features obtained from the detection step. A supervised machine learning algorithm was then implemented on the dataset to predict sleepiness. Logistic regression and Naive Bayes achieved 77% and 79% accuracy in predicting sleepiness, but failed to accurately classify sleepiness based on labels. The KNN algorithm was found to achieve the highest accuracy for sleepiness prediction with an accuracy of 97.5%. It ranked all three labels with higher accuracy and was found to be the best algorithm in the dataset for sleepiness prediction. SVM is second only to KNN, achieving an accuracy of 87%. K-Nearest Neighbors and Support Vector Machines have proven to be the best supervised machine learning algorithms to more accurately classify sleepiness and predict sleepiness. Our approach benefits taxi owners, drivers, students, teachers, staff and organizations who need to monitor their concerns, analyze and review their behavior, which can help them take constructive action to prevent sleepiness, increase the productivity.