Academic • Machine Learning/Healthcare
About The Project
Built a full ML workflow on the Heart Failure Prediction dataset (Kaggle): data cleaning (invalid RestingBP/Cholesterol removal), feature scaling, label/one-hot encoding, and PCA for dimensionality reduction/visualization. Trained and compared Random Forest, KNN, and a PyTorch NN with stratified K-fold CV and early stopping; emphasized reproducible evaluation and feature importance.
Achievements
Random Forest reached 89.33% accuracy; mean ROC-AUC ≈ 0.929 (stratified CV)
Implemented PCA to analyze class separability and guide model selection
Built a clean, modular pipeline for training, tuning, and validation across models
More Projects
Links
sunny.sunho.park@gmail.com
