Analyzing Effective Factors of Online Learning Performance by Interpreting Machine Learning Models

Abstract

Analyzing the effective factors influencing online learning performance is a research topic that has garnered significant attention. Traditional approaches, such as multiple regression and structural equation models, tend to assume linearity, while non-linear machine learning models lack interpretability. To address this gap, we propose a framework that interprets machine learning models to analyze the effective factors of online learning performance. By applying this framework to four benchmark datasets of online learning, we examine the differential impact of various factors on performance, explore the interactions among these factors, and identify the key factors for representative learners. Our findings indicate that: 1) non-linear machine learning models, particularly Decision Regression, offer better representation of the non-linear relationship between effective factors and online learning performance compared to classical multivariate regression; 2) the factor of online learning behavior exerts a greater influence on performance than demographic features, academic background, or online curriculum design; 3) online learning behavior features exhibit additional interaction effects on performance; and 4) learners with medium performance are influenced by diverse effective factors, with active participation in online learning activities emerging as the most crucial means to improve performance. Interpreting machine learning models presents an innovative approach for analyzing the effective factors of online learning performance, which can be extended to other factor analysis studies. The results of this research provide valuable insights for optimizing machine learning models in predicting online learning performance and enhancing learner outcomes.