AI / ML / Personal project
Video Game Hit Predictor
An ensemble model predicting whether games exceed one million sales.
The problem
Reliable classification with imbalanced sales classes.
The approach
Combined Keras DNN and XGBoost with SMOTE, target encoding and F1-optimized thresholds.
Results and scope
- Approximately 84–86% test accuracy and 0.90–0.92 AUC-ROC.