Forecasting PM2.5 Daily Concentration in Baghdad, Iraq Based on Improving Random Forest Algorithm

Forecasting air quality in urban areas is complex due to difficulties in accurately defining emission flux density and the meteorological fields. Combustion gases from human and social activities are the most significant sources of PM 2.5, which is a major air pollutant. Accurate and reliable prediction of PM 2.5 levels is crucial for assessing health risks. Forecasting PM2.5 daily concentration, in general, has been predicted by Random Forest (RF) as a machine learning algorithm. However, the RF performance is highly sensitive to the choice of its hyperparameters, which usually necessitates careful tuning. Consequently, searching for the optimal set of RF hyperparameters constitutes an essential step when attempting to improve model efficiency. Various techniques have come into view for effective hyperparameter tuning of RF. Meta-heuristic optimization methods, with their strong local search abilities, can prevent the training network from getting trapped in local optima and increase the likelihood of identifying the global optimum. This paper proposes employing the Coati Optimization Algorithm (COA), a meta-heuristic approach, to improve RF hyperparameter determination, and, consequently, forecasting PM 2.5 concentrations. Daily PM 2.5 concentrations in Baghdad, Iraq, from 2019 to 2023 are gathered to train RF models and assess the proposed COA performance. The effectiveness of COA is estimated using several metrics. Overall, our proposed COA approach demonstrates superior performance in terms of evaluation criteria compared to other methods in both training and testing daily PM 2.5 concentrations.