Student Performance Predictive Model with Machine Learning

Develop a comprehensive predictive analytics pipeline using scikit-learn that forecasts individual student academic performance across multiple courses. The model should integrate historical grade data, attendance records, learning platform engagement metrics, and demographic information. Create a modular Python script that can handle feature engineering, model training, and real-time prediction with at least 85% accuracy. Include robust error handling, cross-validation techniques, and a mechanism to automatically retrain the model quarterly.