Optimizing Machine Learning Models for Predicting User Engagement in Online Learning Platforms: The Role of Feature Selection

Online education has emerged as a popular alternative to traditional learning environments, driven by its flexibility in terms of time, location, and pace. Understanding the factors that influence user engagement and course completion is essential for designing effective online learning experiences. This study explores how features such as time spent on courses, number of videos watched, quizzes taken, quiz scores, and device type impact engagement and course completion. Using a dataset incorporating these features, the Random Forest algorithm was employed to identify feature importance, highlighting its effectiveness in handling high-dimensional data. Additionally, the performance of five machine learning algorithms—Logistic Regression, Decision Trees, Random Forests, Support Vector Machines (SVM), and Gradient Boosting Machines (XGBoost)—was compared. Results indicate that Random Forests and XGBoost consistently achieve high accuracy, demonstrating robustness to feature reduction. In contrast, Logistic Regression showed sensitivity to reduced feature sets, while Decision Trees and SVM exhibited moderate performance drops. The findings provide valuable insights into optimizing online learning platforms, enhancing student engagement, and reducing dropout rates by identifying key behavioral predictors and selecting suitable algorithms for predictive modeling.