Student Anomaly Detection with MLOps

This project implements an anomaly detection model to identify students at risk of failure. It follows MLOps best practices including experimentation tracking, model versioning, and automated pipelines.

🚀 Features

• Anomaly Detection: Isolation Forest model to detect unusual student behavior.
• Risk Assessment: Hybrid scoring combining ML anomaly scores with rule-based heuristics.
• Experiment Tracking: Weights & Biases (W&B) integration for metric logging.
• Model Registry: Automatic versioning of trained models using W&B Artifacts.
• CI/CD: GitHub Actions pipeline for automated testing and checks.
• API: Flask-based REST API for real-time and batch predictions.

🛠️ Setup

1. Clone the repository

   git clone <repo-url>
   cd anomalydetectionmodel

2. Install dependencies

   pip install -r requirements.txt

3. Set up Weights & Biases

   • Create an account at wandb.ai
   • Login locally:

     wandb login

🏃‍♂️ Training

To train the model and log experiments to W&B:

python train_model.py

This will:

• Load data from data/
• Train the Isolation Forest model
• Log metrics (F1 score, contamination) to W&B
• Save the model to models/ and upload it to W&B Artifacts

🌐 API

Start the Flask API:

python app.py

Endpoints:

• POST /predict: Predict risk for a single student
• POST /predict_batch: Predict risk for a batch of students
• GET /diagnose: Run diagnostic tests

🧪 MLOps Pipeline

The project uses GitHub Actions for CI/CD:

• Linting: flake8 checks for code style issues.
• Testing: pytest runs unit tests to ensure model integrity.
• Trigger: Runs on every push to main and pull requests.