ICU Mortality Prediction Model

This project focused on predicting ICU patient mortality from a Kaggle healthcare dataset. I used Python, Pandas, and PyTorch to clean the data, compare multiple model families, and iterate toward the strongest-performing approach.

Highlights

• Built a full training pipeline for preprocessing, feature comparison, evaluation, and hyperparameter tuning
• Tested multiple architectures including logistic regression, random forest, and neural networks
• Chose the best-performing model based on precision and recall tradeoffs
• Reached approximately 99% accuracy during the final tuned runs

What I learned

Working with medical data made model evaluation more important than headline accuracy. Comparing multiple approaches and understanding where they performed well helped turn this into more than just a training exercise.

Predictive Model AUC Performance