📌 System Overview

• Heavy Usage Prediction (Next 15-min bucket)
• Next Category Prediction (Active Apps Only)
• Memory Windows: 3, 6, 9 buckets
• Best Performing Model: Random Forest

📈 Heavy Usage AUC

🔥 Heavy Burst Recall

⚡ Heavy Usage Ratio per User

🎯 Next Category Prediction Accuracy

🏆 Best Model Comparison

📊 Window Size Comparison

📈 Overall System Performance

🧠 Interpretation

• Heavy usage bursts are highly predictable (AUC up to 0.93).
• Next category behavior can be predicted up to 70%.
• Structured users show stronger behavioral momentum.
• Random Forest performs most consistently.
• Oversampling and undersampling degraded performance.
• The system models behavioral momentum — not random usage.