Weatheria is a machine learning-powered weather prediction platform that leverages advanced algorithms to analyze historical climate data. It predicts Earth surface temperatures based on spatial (latitude, longitude) and temporal (year, month) features, providing insights into climate trends and the impact of global warming.
The project incorporates a highly modular and scalable architecture:
- Frontend: Built with Nuxt.js and styled using TailwindCSS for a modern, responsive, and user-friendly design. Deployed seamlessly on Vercel.
- Backend: Developed using FastAPI, containerized with Docker, and hosted on Google Cloud VM instances. This ensures robust, scalable, and efficient API endpoints for serving predictions.
- Machine Learning: The models were developed using Scikit-learn, employing four algorithms:
- Random Forest: Achieved the highest accuracy with an R² score of 0.9857.
- K-Nearest Neighbors (KNN): Provided reliable predictions, balancing simplicity with effectiveness.
- Support Vector Regression (SVR): Implemented with various kernels to model non-linear relationships.
- Linear Regression: Used as a baseline to compare performance against more advanced algorithms.
- Data Handling: The dataset was sourced from the Berkeley Earth Surface Temperature repository. Rigorous preprocessing steps were applied:
- Removed incomplete records prior to 1870.
- Excluded missing temperature values for data integrity.
- Extracted and scaled features (year, month, latitude, longitude) for optimal model training.
The integration of these technologies and methodologies allowed Weatheria to deliver highly accurate predictions and valuable insights, making it a meaningful contribution to understanding climate change.
