An Innovative Hybrid Model of ARIMA and Random Forest for Predicting MOOC Learning Outcomes

Abstract

With the rapid growth of digital education, MOOCs have emerged as an essential tool for promotiPng lifelong learning among the public. However, pervasive issues of high learner attrition and low completion rates necessitate the establishment of a predictive mechanism capable of forecasting learning risks while providing decision-making support in advance. This study proposes an innovative predictive framework that integrates time series analysis and machine learning to enhance the prediction accuracy and model stability of MOOC learning outcomes. The prediction model process is divided into five stages: data preprocessing, time series forecasting using the ARIMA model, feature engineering and dimensionality reduction through PCA, RF modeling, and model performance evaluation. Initially, the ARIMA model was used to capture the linear trend, followed by RF modeling to fit the nonlinear components of the residuals. The results indicate that the proposed ARIMA+PCA+RF hybrid framework demonstrates significantly superior prediction performance on both the training set (80%) and the testing set (20%). In the final evaluation stage, the hybrid model’s performance indicators on the test data are as follows: F1-score = 0.8961, MSE = 1.912, RMSE = 1.3827, MAE = 1.0243, R² = 0.9024, clearly outperforming the average performance of traditional models (RA, SVM, ARIMA, RF). Particularly in terms of RMSE and the F1-score, the hybrid model consistently ranks highest among all comparative models, achieving an RMSE of 1.13 and an R² of 0.91 in the F1–F14 combination. This hybrid predictive framework enhances the interpretability and predictive accuracy of student performance models, enabling early identification of at-risk learners and facilitating timely intervention.