ML Project

Abstract

Abstract.pdf

Dataset: https://archive.ics.uci.edu/dataset/296/diabetes+130-us+hospitals+for+years+1999-2008

Description

Developed machine learning models to predict 30-day readmissions for diabetes patients using a dataset from 130 U.S. hospitals (1999–2008) with 101,766 records and 48 features. Conducted thorough exploratory data analysis, feature engineering, and data cleaning to enhance model performance.

Key techniques:

• Encoding categorical variables, handling missing values, and reducing irrelevant or noisy features.
• Applied models like Decision Tree, Random Forest, Gradient Boosting, GaussianNB, XGBoost, and Logistic Regression, with SMOTE to address class imbalance.
• Hyperparameter tuning using RandomizedSearchCV and GridSearchCV.

Highlighted techniques:

• Evaluation Metrics: Recall and F1-score for identifying minority class.
• Explainability: SHAP values to improve model transparency.

This project showcases the transformative potential of predictive analytics in healthcare, enabling proactive interventions and reducing readmission rates.

Contents

• notebooks/: Contains the Jupyter notebook with the code implementation.
• docs/: Contains the abstract and the report.

Installation

1. Clone the repository:

   git clone https://github.com/your-username/ML_project.git