🩺 Explainable Diabetes Prediction System

An explainable machine learning–based system for predicting diabetes using clinical and demographic data. The system leverages a Random Forest classifier and SHAP (SHapley Additive exPlanations) to provide both accurate predictions and transparent, human-interpretable explanations for each decision.

📌 Project Overview

Diabetes is a chronic metabolic disease that requires early detection for effective management and prevention of complications. Traditional diagnostic approaches can be time-consuming and dependent on specialist interpretation. This project proposes an automated, data-driven, and explainable system that predicts whether an individual is diabetic or non-diabetic based on routinely collected medical attributes.

Unlike black-box models, this system emphasizes model explainability, allowing healthcare practitioners and users to understand why a prediction was made.

🎯 Aim and Objectives Aim

To develop an explainable machine learning system for accurate prediction of diabetes using clinical data.

Objectives

To analyze diabetes-related clinical data and identify key predictive features

To build a Random Forest–based diabetes prediction model

To evaluate model performance using standard classification metrics

To apply SHAP for explaining global and individual predictions

To deploy an interactive user interface using Streamlit

📊 Dataset

Source: Pima Indians Diabetes Dataset (UCI / Kaggle)

Records: 768

Features: 8 clinical attributes

Target Variable:

0 → Non-Diabetic

1 → Diabetic

Features Used

Pregnancies

Glucose Level

Blood Pressure

Skin Thickness

Insulin

Body Mass Index (BMI)

Diabetes Pedigree Function

Age