Predicting Student Outcomes Using a Pipeline-Driven Deep Neural Network on E-Learning Reaction Data

Due to the swift growth of e-learning platforms, large amounts of data about student interactions and reactions have been generated creating new opportunities for data-driven insights into learning effectiveness and predictive student performance. A structured Deep Learning Pipeline (DLP) is proposed in this work to forecast student outcomes based on behavioral data collected during online classes and student responses. The main idea of this work is to improve performance prediction accuracy and reliability when compared to traditional machine learning models and a basic deep learning model. We developed a systematic pipeline that efficiently and reliably integrates model training hyperparameter optimization feature normalization and data preprocessing. The proposed DLP model is evaluated against several well-established algorithms, including XGBoost Regressor, K-Nearest Neighbors (KNN), Decision Tree, Multi-Layer Perceptron (MLP), Random Forest, Support Vector Machine (SVM), Radial Basis Function (RBF) network, and a simple DNN without pipeline structure. All experiments are implemented using Python libraries, ensuring consistency and reproducibility. The results demonstrate that the DLP-based DNN achieves the highest accuracy of 90% in predicting student performance outcomes, marginally outperforming both the Random Forest and baseline DNN models (each at 89% accuracy). These findings indicate that a structured pipeline significantly enhances the model’s predictive ability by systematically handling data preprocessing, feature engineering, and model optimization steps. The proposed solution offers practical benefits for educators and e-learning platform developers aiming to deliver real-time, data-driven insights into student progress and engagement.