COVID-19 Chest Classification System

Portfolio project

<<<<<<< HEAD

chest_classification_System# chest_classification_System

chest_classification_System

=======

COVID-19 Chest Classification System

A data science project combining chest X-ray image classification (CNN) and clinical vitals classification (Random Forest), served via a FastAPI backend.

Project Structure

Setup (VS Code)

Create and activate a virtual environment:

Install dependencies:

Step 1: Prepare Clinical Data

Run the EDA/preprocessing script to generate data/clean_clinical_data.csv:

Step 2: Download Image Dataset

The chest X-ray dataset (COVID-19 Radiography Database) is on Kaggle:

Get your API key from kaggle.com -> Account -> Create New API Token (kaggle.json)

Place it in ~/.kaggle/kaggle.json (or C:\Users\<you>\.kaggle\kaggle.json on Windows)

Run:

Make sure the resulting folder is data/COVID-19_Radiography_Dataset/ with one subfolder per class (COVID, Normal, Lung_Opacity, Viral Pneumonia), each containing an images/ subfolder.

Step 3: Train Models

Train the clinical model (fast, runs on CPU):

Train the image model (recommend GPU, will be slow on CPU):

Both scripts save trained models/artifacts into models/.

Step 4: Run the API

Then open http://127.0.0.1:8000/docs for the interactive Swagger UI where you can:

POST to /predict/clinical with vitals/symptoms (JSON) -> returns COVID probability

POST to /predict/image with a chest X-ray image file -> returns predicted class + confidence

The API will tell you (at /) which models are currently loaded, so you can run it even if you've only trained one of the two models so far.

Notes & Limitations

The clinical dataset is heavily imbalanced (~1.4% positive), so the clinical model prioritizes recall via class_weight="balanced", but precision remains low. Clinical vitals alone are not a reliable diagnostic signal.

The image and clinical datasets are not paired by patient, so this project presents two complementary models rather than a single fused multimodal model.

This is an educational/portfolio project, not a medical diagnostic tool.