Diabetes-Prediction

Diabetes prediction using data analysis & ML on health metrics — includes cleaning, visualization, feature engineering, and model evaluation.

This repository contains a complete end-to-end data analysis and machine learning project for predicting the likelihood of diabetes using the Pima Indians Diabetes dataset.

📊 Dataset Description

The dataset consists of medical diagnostic measurements for Pima Indian women. It includes 768 rows and 9 columns (8 features + 1 target).

Features:

• Pregnancies
• Glucose
• BloodPressure
• SkinThickness
• Insulin
• BMI
• DiabetesPedigreeFunction
• Age
• Outcome (0 = Non-diabetic, 1 = Diabetic)

📌 Project Steps

1. Data Loading & Cleaning

   • Replaced invalid zeros in Glucose, BloodPressure, SkinThickness, Insulin, and BMI with NaN.
   • Imputed missing values using median strategy.

2. Exploratory Data Analysis (EDA)

   • Boxplots revealed outliers in Insulin, BMI, SkinThickness, and Age.
   • Correlation heatmap showed Glucose had the strongest correlation with diabetes outcome.

3. Feature Scaling & Splitting

   • Applied StandardScaler for normalization.
   • Split dataset into train and test sets (80-20).

4. Model Building & Evaluation

   • Trained Logistic Regression, KNN, and Random Forest.
   • Random Forest performed best with ~78% accuracy.
   • Evaluation metrics included confusion matrix and classification report.

5. Feature Importance

   • Top predictors: Glucose, BMI, DiabetesPedigreeFunction, and Age.

✅ Conclusion

This project provides a solid baseline for diabetes prediction using traditional machine learning models. Further improvement can be done via hyperparameter tuning or advanced algorithms.

📁 Files

• diabetes_prediction.ipynb – Jupyter Notebook with full workflow.
• diabetes.csv – Dataset file used in the project.
• README.md – Project overview and documentation.

---

🔗 Author

Abhishek Kunbhare