🩺 Medical Machine Learning: Pneumonia Diagnosis Pipeline

An end-to-end comparative study evaluating supervised learning classifiers, soft-voting ensembles, and computer vision on clinical feature metrics and raw chest X-ray images.

---

📌 Project Overview

This repository investigates the practical application of Machine Learning in diagnosing Pneumonia across two distinct modalities:

1. Tabular Feature Analysis: Comparative benchmarking of standalone supervised algorithms and multi-tiered Soft-Voting Ensembles built on clinical feature data.
2. Computer Vision Classification: Automated preprocessing and evaluation of raw 2D chest X-ray images using flattened Support Vector Machine (SVM) classification.

Special emphasis was placed on mitigating class imbalance (class_weight='balanced') to optimize diagnostic recall and minimize critical False Negatives (mislabeled patient diagnoses).

---

🛠️ Key Technical Highlights

• Automated Pipelines: Built robust sklearn.pipeline.Pipeline chains combining StandardScaler with classifier models to prevent data leakage during evaluation.
• Class Imbalance Mitigation: Integrated balanced class weighting across Decision Trees, Logistic Regression, and SVMs to improve recall on the minority class (0.0 / No Pneumonia).
• Multi-Level Soft-Voting Ensembles: Constructed hierarchical ensembles combining high-performing base models (Random Forest, Logistic Regression, SVM) using soft probability weighting.
• Computer Vision Pipeline: Preprocessed raw X-ray scans via grayscale conversion (rgb2gray), 128x128 dimensional resizing, 1D array flattening, and standard scaling.
• Rigorous Cross-Validation: Applied 5-Fold Stratified K-Fold Cross-Validation (StratifiedKFold) across all non-testing partitions to ensure generalization.

---

📊 Benchmark Model Performance

1. Tabular Clinical Feature Models (Dataset: 583 Observations)

Model Architecture	Hyperparameters / Preprocessing	Key Metrics (Test / K-Fold CV)	Primary Strength
Logistic Regression	liblinear, max_iter=150, Balanced	73.1% CV Accuracy (71.6% Test)	Strong overall generalization; robust recall
Support Vector Machine (SVM)	kernel='rbf', gamma=5, Balanced	71.1% CV Accuracy (71.6% Test)	Effective margin separation for complex features
Random Forest Ensemble	n_estimators=150, max_depth=5, Balanced	69.2% CV Accuracy (67.2% Test)	Reduced overfitting compared to single decision trees
Optimal Hierarchical Ensemble	Soft-Voting (Random Forest + SVM + LogReg)	69.8% CV Accuracy (64.7% Test)	Best Minority Class Recall & F1-Score
K-Nearest Neighbors (KNN)	n_neighbors=7	63.4% CV Accuracy (64.7% Test)	Good positive class detection
Decision Tree Classifier	max_depth=5, Balanced	60.4% CV Accuracy (60.3% Test)	High interpretability
Gaussian Naive Bayes	var_smoothing=1e-8	54.1% CV Accuracy (59.5% Test)	Fast baseline execution

2. Computer Vision Pipeline (Raw Chest X-Ray Images)

• Architecture: Support Vector Classifier (kernel='linear', probability=True)
• Preprocessing: Grayscale transformation, 128x128 spatial resizing, standard scaling
• Test Accuracy: 75.0%
• Pneumonia Sensitivity (Recall): 0.99 (99%)

---

🏗️ Repository & Directory Structure

├── MLPneumonia.py             # Primary Python script / Colab notebook source
├── portfolio_thumbnail.png    # Project visual graphic for portfolio summary
├── pneumonia_raw.csv          # Tabular clinical dataset (raw)
├── chest_xray/                # Raw X-ray image directory (train / val / test)
└── README.md                  # Project documentation

---

🚀 How to Run Locally

1. Clone the Repository

git clone https://github.com/your-username/your-repo-name.git
cd your-repo-name

2. Install Dependencies

pip install pandas numpy scikit-learn scikit-image seaborn matplotlib

3. Run the Pipeline

python3 MLPneumonia.py

---

Developed as part of an academic machine learning investigation into AI-assisted medical diagnostics.