Weather Trend Forecasting

Advanced weather trend forecasting project using the Global Weather Repository dataset.

Project Files

• weather_trend_forecasting_advanced.ipynb - main notebook with data loading, cleaning, EDA, anomaly detection, forecasting models, ensembles, and feature importance.
• data/GlobalWeatherRepository.csv - weather dataset used by the notebook.
• requirements.txt - Python packages needed to run the notebook.

Project Scope

• Data cleaning and preprocessing: timestamp parsing, duplicate handling, explicit feature selection, scaling, categorical encoding, and conservative outlier handling.
• EDA: temperature, precipitation, correlation/association tables, weather condition summaries, time patterns, air quality relationships, and geography/spatial plots.
• Advanced EDA: IQR outliers, location-relative z-score anomalies, and Local Outlier Factor anomaly detection.
• Forecasting: Ridge Regression, Random Forest, HistGradientBoosting, soft average ensemble, and holdout stacking ensemble.
• Evaluation: MAE, RMSE, R2, and forecast accuracy within +/-2C, +/-3C, and +/-5C.
• Unique analyses: climate/time patterns, environmental impact, feature importance, spatial/geographical patterns.
• PM Accelerator mission statement: included in the notebook with source attribution.

How To Run

1. Create and activate a Python environment.
2. Install dependencies:

pip install -r requirements.txt

3. Open and run:

jupyter notebook weather_trend_forecasting_advanced.ipynb

The notebook expects the dataset at data/GlobalWeatherRepository.csv.

Current Best Model

Strict future-holdout validation selects Random Forest trained on the conservative filtered training set:

• MAE: 3.057 C
• RMSE: 4.464 C
• R2: 0.831
• Accuracy within +/-2C: 50.4%
• Accuracy within +/-3C: 64.1%
• Accuracy within +/-5C: 81.0%