Hospital Readmission Prediction for Diabetes Patients

Overview

This repository contains the Python code for predicting hospital readmission within 30 days for patients diagnosed with diabetes, based on data from 130 US hospitals covering ten years (1999-2008). The project is divided into two main parts: building a basic predictive model and developing an improved model using classification and clustering techniques.

Objective

The goal is to accurately predict early hospital readmissions (<30 days) of diabetes patients using machine learning models. The project aims to showcase data preprocessing, visualization, model building, evaluation, and enhancement through clustering.

Data

The dataset, diabetic_data.csv, consists of 47 features and 101766 instances representing hospital records, laboratory medications, and patient stays up to 14 days.

File Descriptions

• basicModel.py: Contains the implementation of the basic predictive model. This script demonstrates data cleaning, transformation, feature selection, model training (using a linear model), and evaluation with a focus on handling missing values aggressively and dealing with imbalanced data through a sampling technique.

• improvedModel.py: Implements the improved model with a comprehensive approach towards data cleansing, feature engineering, and model tuning. It introduces cluster-based classification using the K-Means algorithm to enhance model performance and provides a detailed comparison between the basic and improved models.

Usage

Data Preprocessing

1. Aggressive cleaning: Removing columns with high missing values, replacing '?' with NaN, and dropping near-zero variance columns.
2. Feature engineering: Converting the 'readmitted' column into a binary feature for prediction.
3. Data balancing: Techniques like oversampling to balance the dataset.

Model Building

• Basic Model: A linear model such as Logistic Regression is used focusing on selected predictors, evaluated using cross-validation and performance metrics suitable for imbalanced datasets.
• Improved Model: Explores advanced preprocessing, utilizes the entire dataset, and implements cluster-based classification for enhanced performance.

Evaluation

Models are evaluated based on accuracy, precision, recall, F1-score, and the area under the ROC curve. The improved model's performance is also compared against the basic model to highlight the effectiveness of the clustering approach.

How to Run

Ensure you have Python 3 and all necessary libraries installed. Run the following commands in your terminal:

pip install -r requirements.txt
python basicModel.py
python improvedModel.py