🛰️ AI-Powered Meteorological Prediction Engine

Localized Machine Learning Forecasting Station for Metuchen, NJ

1. The High-Level Architecture

The program is built on a Three-Tier Architecture that ensures the system remains autonomous and self-correcting:

Data Tier (Ingestion): Connects to Open-Meteo API using requests-cache. It fetches both 3 years of Historical "Truth" and a 10-day "Future" forecast.
Logic Tier (Intelligence): Utilizes a dual-model approach:
- Random Forest Regressor: Predicts the intensity of precipitation.
- Logistic Regression: Predicts the probability of rain/snow events.
Presentation Tier (UI): A Flask-based web server that renders dynamic Matplotlib charts and customizes units for the end-user.

2. The Data Flow & Logic Engine

Stage	Process	Technology
Training	Learns local patterns (Temp, Humidity, Pressure) from 2023–2025.	scikit-learn
Inference	Applies learned patterns to the 240-hour forecast window.	Random Forest
Transformation	Calculates Snow Ratio (10:1) and converts Metric to Customary units.	Pandas / NumPy
Validation	Logs "Predicted vs. Actual" to SQLite for error tracking.	SQLite3

3. Key Design Features (The "Why")

Lag Features: The model uses "Weather Memory"—it looks at the conditions from one hour ago to understand the trajectory of a storm, not just a single snapshot.
Snow-Switch Logic: A software "gate" monitors temperature; at < 33.8°F, it triggers the snow accumulation algorithm.
Self-Correction: By calculating the Mean Absolute Error (MAE), the system provides a transparency score, telling the user exactly how much to trust the model today.

4. Technical Stack Summary

Languages: Python 3.11
Database: SQLite3
Deployment: Debian Linux virtual machine
Web Application Framework: Flask

💡 Feedback Loop The program doesn't just "guess" and forget; it writes its guess to a database and checks back later to see if it was right. This process of continuous monitoring and logging is the essence of Machine Learning Operations (MLOps).