Predicting Engagement and Boredom in Online Video Lectures

The study uses specific datasets from Daisee (which features videos of many people) and Kaggle (which features photos of a variety of people) to examine the unique issues of forecasting student interest and boredom in online lectures. Convolutional neural networks (CNN) was used to retrieve image features. For analysing sequential data collected from image frames of videos, long short-term memory (LSTM) was utilised. This is a hybrid technique. This approach makes it possible to independently evaluate how well each algorithm captures student reactions in online learning settings. Three datasets: Daisee, Kaggle, and an aggregated one, were used to train and assess predictive models. While LSTM concentrates on evaluating sequential information from video footage which is converted to image frames during input, CNN extracts features from images. The comparison research sheds light on how differently each algorithm performs in terms of engagement and boredom prediction across various datasets. Giving teachers useful resources to evaluate and improve their online teaching approaches is the main goal. Instructors can adjust their teaching strategies based on an intricate understanding of students' responses by predicting students' interest levels using both visual and sequence data. This increases student engagement and reduces the risk of boredom. The study's conclusions give educators practical advice that will make it easier to modify lesson plans for a more successful and interesting online learning environment.