Hospital Readmission Prediction Model

Overview

This is my project for predicting hospital readmissions using machine learning. The focus is on creating a binary classification model that is both effective and user-friendly. I used LightGBM as the main algorithm because it’s fast and performs well on a variety of datasets. To handle class imbalance, I included SMOTEENN, and for fine-tuning the model, I used GridSearchCV. There are also visualizations, like an ROC curve, to help understand the model's performance.

What It Does

Data Preprocessing

• It handles missing values by filling numeric ones with the median and encoding categorical data into dummy variables.
• Makes sure all data is numeric, replacing any leftover missing values with zeros.

Model Training

• Trains a LightGBM model, which is optimized using GridSearchCV.
• Uses SMOTEENN to balance the dataset by over-sampling the minority class and cleaning noisy data.
• Tunes parameters like num_leaves, max_depth, and learning_rate to get the best results.

Evaluation and Visualization

• Plots an ROC curve to show how well the model distinguishes between readmitted and non-readmitted patients.
• Shows accuracy for predicting 'Unadmitted' patients through a bar chart.

How to Set It Up

1. Clone the repository.

2. Run the setup script:

   bash setup.bash
   

Notes

• This project helped me practice working with machine learning, imbalanced datasets, and automating cross-platform setups. If you encounter any issues, re-running setup.bash should resolve missing dependencies. The project has been tested on macOS and Linux (Ubuntu).