Comeback Prediction in Professional Esports (League of Legends)
Python, scikit-learn, Plotly, pandas, Git
- Built a Random Forest classification model to predict win-from-deficit ("comeback") outcomes using 150K+ professional esports match records and early-state features at a fixed timestamp.
- Cleaned and prepared large-scale match data; created a reproducible dataset with engineered features capturing resource/experience gaps, objective control, and engagement efficiency.
- Addressed class imbalance (~16%) by optimizing for F1-score, applying class_weight, and tuning hyperparameters using GridSearchCV (k-fold cross-validation).
- Improved F1-score by 347% over baseline Decision Tree model through feature engineering and model selection; communicated tradeoffs using precision/recall rather than accuracy.
- Validated statistical claims using permutation testing and evaluated fairness across regions to confirm no meaningful performance disparity across cohorts.