Hybrid Machine Learning-Based Air Quality Forecasting Using CatBoost with Hunger Games Search and Arithmetic Optimization Algorithm

Abstract

Air pollution is a significant global concern, posing a major challenge to sustainable development if neglected. Leveraging mathematical frameworks through ML offers an optimal and cost-effective solution for modeling air pollution. This investigation introduces hybrid ML-based frameworks to anticipate air quality pollutants and classify air quality. Specifically, the CatBoost algorithm was combined with the Arithmetic Optimization Algorithm (AOA) and the Hunger Games Search algorithm (HGS) for prediction and classification purposes. The database comprises daily time series data of air pollutants in China from 2018 to 2021. Autocorrelation function (ACF) and partial autocorrelation function (PACF) approaches were utilized to select input combinations for each pollutant. Results indicate that the integrated model provides highly accurate forecasts of pollution index time series using the regression method. Furthermore, evaluation metrics reveal that the classification method surpasses the regression method regarding accuracy for predicting the AQI.