🌟 PPG Blood Glucose Diabetes Classification

Welcome to the PPG Blood Glucose Diabetes Classification project! 🎉
This repository provides a pipeline for estimating blood glucose levels using Photoplethysmography (PPG) signals.
By combining advanced signal processing, machine learning, and visualizations, we enable non-invasive diabetes screening for modern preventive healthcare. 🩺💡

---

📋 Project Overview

This project uses PPG—an optical technique for capturing blood volume changes—to predict blood glucose levels non-invasively.

The modular framework integrates signal engineering and machine learning, making it suitable for experimentation and real-world use.

✨ Key Features

• 📊 Data Processing: Preprocess raw PPG segments and extract physiological features
• 🤖 Machine Learning Models: Random Forest, Gradient Boosting, SVM, LightGBM, Logistic Regression, and Ensemble Methods, Stacking and Voting Classifiers
• 📈 Visualizations: ROC curves, confusion matrices, and feature importance plots
• 🧪 Evaluation: Subject-wise StratifiedGroupKFold cross-validation to prevent data leakage and ensure real-world applicability

---

📂 Project Structure

PPG_Blood_Glucose_JB_Implementation/
├── datasets/
│   ├── ppg_bagging_tree_features.csv
│   ├── ppg_specific_features.csv
│   ├── processed_metadata.csv
│   ├── PPG-BP.xlsx
│   └── 0_subject/
├── models/
│   ├── random_forest.pkl
│   ├── svm.pkl
│   └── ...
├── outputs/
│   ├── confusion_matrix_randomforest.png
│   ├── roc_randomforest.png
│   └── ...
├── src/
│   ├── excel_handling.py
│   ├── data_preprocessing.py
│   ├── train_models.py
│   └── evaluate.py
├── requirements.txt
└── README.md

---

🚀 Getting Started

Let’s get you up and running in no time!

✅ Prerequisites

• Python 3.8+ 🐍
• pip (Python package manager)
• Git 📦

---

🛠 Setup Instructions

1. Clone the Repository

git clone https://github.com/Spidy104/PPG_DIABETES_CLASSIFICATION
cd PPG_DIABETES_CLASSIFICATION

2. Set Up a Virtual Environment (Recommended)

python -m venv venv

3. Activate the Virtual Environment

• On macOS/Linux:

  source venv/bin/activate

• On Windows:

  venv\Scripts\activate

4. Install Dependencies

pip install -r requirements.txt

---

🧠 Usage Guide

📁 Step 1: Prepare the Data

Ensure the datasets/ folder contains:

• ppg_bagging_tree_features.csv — Contains extracted features from PPG signals using bagging tree methods for model training.
• ppg_specific_features.csv — Includes domain-specific physiological features derived from PPG signals.
• processed_metadata.csv — Metadata for each sample, such as subject IDs, timestamps, and labels (e.g., glucose levels).
• PPG-BP.xlsx — Raw and reference data, including PPG signals and corresponding blood pressure/glucose measurements.
• 0_subject/ (Raw PPG signals by subject) — Directory with raw PPG signal files, organized per subject for preprocessing.

⚙️ Step 2: Process and Preprocess Data

2.1 Process Excel Metadata

python src/excel_handling.py

2.2 Preprocess Raw Data

python src/data_preprocessing.py

2.3 Feature Extraction

python src/feature_extraction.py

🏋️ Step 3: Train the Models

Run the following command to train all machine learning models:

python src/train_models.py

This will generate model files in the models/ directory:

models/
├── random_forest.pkl
├── svm.pkl
├── gradient_boosting.pkl
├── lightgbm.pkl
├── logistic_regression.pkl
├── stacking_classifier.pkl
└── voting_classifier.pkl

📊 Step 4: Evaluate and Visualize Results

Evaluate the trained models and generate visualizations:

python src/second_model.ipynb

Results and plots will be saved in the outputs/ directory:

outputs/
├── gradient_boosting.jpg
├── LightGBM.jpg
├── Stacking_Classifier.jpg
├── Voting_Classifier.jpg
├── Model_performance.jpg
├── ROC_curves.jpg

---

✅ Example Outputs

Here’s a sneak peek at the insights you'll get:

🧾 Classification Report

Below are the classification reports for each model (Model 2):

Random Forest

Class	Precision	Recall	F1-Score	Support
0	0.830	0.990	0.900	181
1	0.000	0.000	0.000	38
Accuracy			0.820	219
Macro Avg	0.410	0.500	0.450	219
Weighted Avg	0.680	0.820	0.750	219

SVM

Class	Precision	Recall	F1-Score	Support
0	0.830	1.000	0.910	181
1	0.000	0.000	0.000	38
Accuracy			0.830	219
Macro Avg	0.410	0.500	0.450	219
Weighted Avg	0.680	0.830	0.750	219

Gradient Boosting

Class	Precision	Recall	F1-Score	Support
0	0.830	0.980	0.900	181
1	0.330	0.050	0.090	38
Accuracy			0.820	219
Macro Avg	0.580	0.520	0.490	219
Weighted Avg	0.740	0.820	0.760	219

LightGBM

Class	Precision	Recall	F1-Score	Support
0	0.830	1.000	0.910	181
1	0.000	0.000	0.000	38
Accuracy			0.830	219
Macro Avg	0.410	0.500	0.450	219
Weighted Avg	0.680	0.830	0.750	219

Stacking Classifier

Class	Precision	Recall	F1-Score	Support
0	0.840	0.970	0.900	181
1	0.440	0.110	0.170	38
Accuracy			0.820	219
Macro Avg	0.640	0.540	0.540	219
Weighted Avg	0.770	0.820	0.770	219

Voting Classifier

Class	Precision	Recall	F1-Score	Support
0	0.830	0.990	0.900	181
1	0.000	0.000	0.000	38
Accuracy			0.820	219
Macro Avg	0.410	0.500	0.450	219
Weighted Avg	0.680	0.820	0.750	219

📊 Gradient Boosting Performance

🧮 LightGBM Confusion Matrix

🤖 Stacking Classifier Results

🗳️ Voting Classifier Results

📈 Model Performance Comparison

🏅 ROC Curves for All Models