diabetes-prediction-analysis

End-to-end analysis and prediction of diabetes risk using the Pima Indians dataset.

Metabolic Health Explorer: Pima Indians Diabetes Analysis

Project Overview

This project analyzes the Pima Indians Diabetes Dataset to identify key indicators of diabetes risk. It progresses from Exploratory Data Analysis (EDA) to building predictive models using jupiter notebook.

Key Insights

• Data Integrity: Identified that the dataset used 0 as a placeholder for missing values in biologically impossible columns (Glucose, Blood Pressure, BMI).
• Correlation: Found strong correlations between Glucose levels and Diabetes outcome.
• Outliers: Handled missing data by imputing median values to preserve distribution integrity.

Tech Stack

• Python: Pandas for data manipulation.
• Visualization: Seaborn & Matplotlib for heatmaps and distribution plots.
• Machine Learning: Scikit-Learn (Linear Regression, Random Forest, Logistic Regression).

How to Run

1. Clone the repository.
2. Install dependencies: pip install pandas seaborn scikit-learn.
3. Run the script metabolic_health_eda.py.

Model Performance (Logistic Regression)

We trained a Logistic Regression model to predict diabetes onset based on diagnostic measures.

• Overall Accuracy: 75.32%
• Key Insight: The model is highly effective at identifying healthy patients (Precision: 0.80) but requires further tuning to improve sensitivity for detecting positive diabetic cases (Recall: 0.62).

Confusion Matrix Results

	Predicted Healthy	Predicted Diabetic
Actual Healthy	High Accuracy	Low False Positives
Actual Diabetic	Moderate False Negatives	Moderate True Positives

Patient Segmentation (Unsupervised Learning)

To understand patient profiles beyond simple "Sick/Healthy" labels, we applied K-Means Clustering (k=3) to segment the population.

Key Findings:

1. Cluster 0 (Metabolic Syndrome): Young patients (avg age 29) with severe obesity (BMI 39) and highest insulin resistance. This group represents the highest intervention priority.
2. Cluster 1 (Older Mothers): Older patients (avg age 45) with a history of multiple pregnancies (avg 7.3). Diabetes risk here is likely age-related.
3. Cluster 2 (Healthy Baseline): Young patients with lower BMI and normal glucose levels. Only 13% of this group is diabetic.