Improving Online Learning Outcomes: A Novel Approach to Detecting Drowsiness and Sustaining Engagement

Abstract

In today's digital learning era, a common yet significant challenge emerges: participants often fall asleep during lecture videos, leading to missed concepts and overwhelming frustration as they struggle to find where they left off. This research delves into a solution designed to enhance comprehension and ease the mental stress associated with these interruptions. Our journey started with data collection from the Internet focusing on Indian and Malaysian contexts we categorized the data into two states active and drowsy, with this dataset in hand we proceeded to a meticulous preprocessing phase, and data augmentation was employed to enhance the diversity of our dataset while image normalization standardized the inputs data balancing was meticulously implemented to provide an unbiased representation of both classes to our model We then set the stage for intense competition among three advanced models: YOLO-V8, DenseNet201, and ResNet152. Each model underwent a preliminary evaluation over 10 epochs, where YOLO-V8 emerged as the frontrunner with a compelling accuracy of 92% on test data. This promising result spurred us to push further, training YOLO-V8 over two extended phases. This paper chronicles the path from identifying a widespread issue to developing a solution that enhances educational comprehension and reduces participant stress. By incorporating the YOLO-V8 model into educational platforms, we introduce an innovative method to detect drowsiness and sustain engagement, ensuring every participant remains engaged in the digital classroom.