Hybrid feature optimization – machine learning approach for predicting particulate matter

Dust storms in arid regions, particularly in the Middle East and UAE, carry substantial particulate matter (PM) over large distances, posing health risks, environmental concerns, and economic challenges. Predicting PM concentrations from dust storms is essential for public health and environmental stability. This study aims to introduce a hybrid feature optimization– machine learning (FO-ML) framework that leverages spatially distributed climatic data, including temperature, relative humidity, atmospheric pressure, wind direction, and wind speed. Data was collected from the Fujairah Research Center and optimized through a sequence of FO preprocessing steps, including normalization, feature selection using Principal Component Analysis (PCA), and analysis with Pearson’s correlation matrix. The predictive ML model was developed using an Artificial Neural Network (ANN), optimized via grid search over the number of hidden layers and neurons. The best-performing configuration for PM2.5 achieved an R² of 0.90, an RMSE of 0.02, and a minimum prediction time of 19.1 seconds; for PM10, R² of 0.88, RMSE of 0.02, and 18.6 seconds. This setup used hidden layers with up to 100 neurons and ReLU activation. This research advances PM predictions, providing a tool for public health, environmental management, and policy decisions to mitigate PM impacts on ecosystems in affected regions.