Construction and Application of a Learning Motivation Decline Prediction Model Based on Educational Big Data

With the continuous expansion of online education, learning motivation decline has emerged as a critical issue affecting learning outcomes, while traditional assessment methods struggle to meet real-time monitoring demands. This research constructs a predictive model for learning motivation decline based on educational big data. Using the PySpark framework to process the Open University Learning Analytics Dataset (OULAD), we achieve precise prediction through a three-stage pipeline: feature engineering, model training, and system development. The feature engineering phase establishes a multi-dimensional feature system encompassing behavioral features, temporal features, and derived features. In model training, Gradient Boosted Trees (GBT) significantly outperform Random Forest (RF) with an AUC of 0.8996 versus 0.8758. The developed early-warning system supports dynamic risk-threshold adjustment and real-time intervention. Empirical analysis reveals that behavioral features contribute 85% of predictive power, with 86.8% of high-risk students having click counts <100 and 52.4% scoring below 75 points. The system response time is <1.5 seconds. The outcomes provide technical support for risk early-warning in online education and hold significant implications for optimizing pedagogical intervention strategies.