portfolio

Weather Prediction ML Model

A machine learning project that predicts weather conditions using historical weather data. Built with scikit-learn, featuring multiple ML models and comprehensive evaluation metrics.

Features

• Multiple ML Models:
  • Random Forest Classifier
  • Logistic Regression
  • Support Vector Machine
  • Gradient Boosting
• Feature Engineering: Advanced preprocessing and feature extraction
• Model Evaluation: Comprehensive metrics including accuracy, precision, recall, F1-score
• Cross-validation: K-fold cross-validation for robust model evaluation
• Visualization: Feature importance plots and confusion matrices
• Model Persistence: Save and load trained models

Technologies Used

• Python 3.8+
• scikit-learn - Machine learning algorithms
• pandas - Data manipulation
• numpy - Numerical computing
• matplotlib & seaborn - Visualization
• joblib - Model serialization

Installation

pip install -r requirements.txt

Usage

Train the model:

python train.py

Make predictions:

python predict.py --temperature 25 --humidity 65 --pressure 1013 --wind-speed 15

Evaluate models:

python evaluate.py

Dataset Features

The model uses the following features for prediction:

• Temperature (°C)
• Humidity (%)
• Pressure (hPa)
• Wind Speed (km/h)
• Wind Direction (degrees)
• Cloud Cover (%)
• Visibility (km)

Target Classes

• Sunny
• Cloudy
• Rainy
• Stormy
• Snowy

Model Performance

Best performing model: Random Forest Classifier

• Accuracy: ~85%
• Cross-validation score: 83%
• Training time: <2 seconds

Project Structure

• train.py - Model training script
• predict.py - Prediction interface
• evaluate.py - Model evaluation script
• models/ - Saved trained models
• data/ - Dataset storage
• src/ - Source modules
  • preprocessing.py - Data preprocessing
  • model_builder.py - Model definitions
  • evaluator.py - Evaluation functions

Future Enhancements

• Deep learning models (LSTM, CNN)
• Real-time weather data integration
• Web API for predictions
• Time series forecasting
• Ensemble methods