Diabetes Prediction using Logistic Regression

This project demonstrates how AI and Logistic Regression can predict Type 2 Diabetes using sparse Electronic Health Record (EHR) data — particularly focused on developing nations where digital health data is limited.

Overview

The project uses the DiaHealth Bangladesh dataset (2024) with 5,437 patients. It handles missing data, scales biometric features, and classifies patients as diabetic or non-diabetic using logistic regression.

Model Performance:

• Accuracy: 77.94%
• ROC AUC: 0.812
• Sensitivity: 65.6%
• Specificity: 78.7%

Tech Stack

• Python 3.13
• Pandas
• NumPy
• scikit-learn

Features

• Handles missing data using median imputation
• Scales features for balanced performance
• Accepts user input for real-time diabetes prediction
• Works efficiently on small or incomplete datasets

How to Run

# 1️ Install dependencies
pip install -r requirements.txt

# 2️ Run the model
python predict_diabetes.py