Data Quality & Leakage in Predictive Analytics | Chapter2 | Yarmouk University د علاء حمود

In this lecture, we cover two of the most practically important topics in predictive analytics — data quality and data leakage. Poor data quality doesn't just reduce model accuracy; it produces models that make confidently wrong predictions. Leakage is even more dangerous: it makes a model look excellent in testing and fail completely in production. Topics covered: The three types of missingness (MCAR, MAR, MNAR) and how to treat each Outliers: the difference between errors and legitimate extremes The IQR rule for outlier detection and how the multiplier affects your training data Shifting label definitions and why they silently corrupt your model The three types of data leakage: target leakage, temporal leakage, and train-test contamination How to build a reproducible, time-based train/validation/test split All examples use the Yarmouk University accounts receivable dataset. 📚 Course: Predictive Analytics for Decision Making 🎓 Instructor: Dr. Aladeen Hmoud 🏛 Yarmouk University