zaheerjk/Thunderstorm-Forecasting-with-MLFlow-Tracking

---

⛈️ Thunderstorm Forecasting with MLflow Tracking

An end-to-end machine learning system for thunderstorm forecasting, combining predictive modeling with MLflow experiment tracking to build, evaluate, and manage machine learning experiments.

The project covers the complete ML workflow from data preparation and feature engineering to model training, hyperparameter tuning, experiment tracking, and deployment.

🚀 Key Features

• ⛈️ Thunderstorm occurrence prediction
• 🌦️ Weather data preprocessing
• 📊 Exploratory data analysis
• 🔧 Feature engineering
• 🤖 Machine learning models
• ⚙️ Hyperparameter tuning
• 📈 Model evaluation
• 🧪 MLflow experiment tracking
• 📦 Model versioning
• 🌐 Deployment

🖼️ Project Preview

🏗️ System Architecture

🧠 ML Pipeline

Weather Data
     ↓
Data Validation
     ↓
Data Preprocessing
     ↓
Feature Engineering
     ↓
Train / Test Split
     ↓
Model Training
     ↓
Hyperparameter Tuning
     ↓
MLflow Experiment Tracking
     ↓
Model Evaluation
     ↓
Model Registry
     ↓
Deployment
     ↓
Thunderstorm Prediction

📋 Model Details

Property	Details
Task	Weather / Thunderstorm Prediction
Domain	Weather Forecasting
Data Type	Weather / Tabular Data
Framework	Scikit-learn
Experiment Tracking	MLflow
Optimization	Hyperparameter Tuning
Deployment	Web Application / API

🧪 MLflow Tracking

MLflow is used to track the machine learning lifecycle, including:

• Model parameters
• Evaluation metrics
• Training experiments
• Model artifacts
• Experiment runs
• Model versions

This makes experiments reproducible, comparable, and easier to manage.

📊 Workflow

1. Load weather data.
2. Validate and preprocess the dataset.
3. Perform exploratory data analysis.
4. Engineer relevant weather features.
5. Train machine learning models.
6. Tune model hyperparameters.
7. Track experiments using MLflow.
8. Compare model performance.
9. Save the best-performing model.
10. Deploy the prediction pipeline.

📤 Output

The system predicts the likelihood of thunderstorm occurrence:

Thunderstorm Prediction: Yes / No
Probability: <VALUE>

The prediction should be treated as a machine learning forecast, not an official meteorological warning.

💻 Run Locally

git clone https://github.com/mdzaheerjk/Thunderstorm-Forecasting-with-MLFlow-Tracking.git

cd Thunderstorm-Forecasting-with-MLFlow-Tracking

pip install -r requirements.txt

mlflow ui

streamlit run app.py

🛠️ Tech Stack

Python • Pandas • NumPy • Scikit-learn • MLflow • Matplotlib • Seaborn • Streamlit

⚠️ Limitations

Forecasting performance depends on the quality, coverage, and geographical diversity of the weather data.

Real-world weather systems are highly dynamic, and predictions may be affected by unseen atmospheric conditions, data drift, and regional differences.

This project should not replace official meteorological forecasting or emergency warning systems.

🔮 Future Improvements

• Real-time weather data integration
• Advanced time-series models
• LSTM / GRU forecasting
• Transformer-based weather models
• Automated model retraining
• MLflow Model Registry integration
• Model monitoring and drift detection
• Real-time weather alerts
• Geospatial thunderstorm prediction

👨‍💻 Author

Md Zaheer JK

AI/ML • Deep Learning • Generative AI • Computer Vision • NLP • MLOps

GitHub: https://github.com/mdzaheerjk Hugging Face: https://huggingface.co/zaheerjk

📜 License

MIT License

---

⛈️ From Weather Data to Actionable Thunderstorm Predictions