Hybrid Forecasting for Air Quality Data

This script performs hybrid forecasting on the UCI Air Quality dataset using:

• Rolling mean naive forecast for stable columns.
• Last observed value for volatile columns (e.g., RH, NOx sensors).

Features

• Preprocessing with forward and backward fill.
• Feature engineering (hour, day, etc.)
• Stationarity check (ADF) [optional].
• RMSE evaluation on last 10% of the dataset.
• Forecasting next 48 hours of air quality metrics.

Requirements

• pandas
• numpy
• matplotlib
• seaborn
• statsmodels
• scikit-learn

Usage

1. Place the AirQualityUCI.xlsx in the same directory.
2. Run hybrid_forecast_main.py.

The script prints:

• RMSE on the last 10% of data for each variable.
• The 48-hour naive forecast values.

Notes

• Columns with high volatility are handled using last-value prediction.
• Other columns use a rolling mean (window size = 5).
• Adjust the window size and volatile column list as needed.