Comparative analysis of regression and enhanced label propagation approaches for predicting airborne dust particle levels in environmental data

Pollution of airborne dust particle poses serious challenges to the environment and human health, and the operational reliability of precision measurement laboratories, particularly in regions characterized by harsh climatic conditions. Accurate prediction of dust particle concentrations remains challenging due to complex nonlinear interactions among meteorological factors and the limited availability of fully labeled environmental datasets. Therefore, there is a critical need for advanced and robust modeling approaches that can improve the prediction accuracy while handling data uncertainty and sparsity. In this article, several regression analysis (models) and label propagation approaches have been applied and examined for predicting the airborne dust particle concentrations within the national measurement & calibration center (NMCC), saudi standards metrology and quality organization (SASO), Riyadh, Saudi Arabia. Six different models, namely, random forest regression (RFR), K-nearest neighbours regression (KNNR), cosine similarity-based label propagation regression (CS_LPR), adaptive fuzzy entropy-based label propagation regression (AFE_LPR), random forest-based label propagation (RF_LPR), and KNN-based label propagation (KNN_LPR) models were developed for forecasting the airborne dust particle levels and investigated the performance of each model. The airborne dust particles were experimentally counted using an advanced particle measurement technique by considering various factors, namely, air quality, environmental temperature, humidity, wind speed, and rainfall. The performance of the developed models was checked using different metrics such as mean absolute error (MAE), mean squared error (MSE), root mean squared error (RMSE), R 2 , adjusted R 2 , mean absolute scaled error (MASE), and Huber loss. The results obtained from each model demonstrate that the label propagation models (CS_LPR, AFE_LPR, RF_LPR, and KNN_LPR) have exhibited well fitted one, and achieved excellent accuracy on the test data (R 2 > 0.98) due to effective learning of the training data, including noise and specific patterns present in the dataset. The finding obtained through this research work emphasize that the label propagation methods can effectively address the prediction challenges in environmental monitoring tasks. This paper addresses the comparative performance and features of each approach in airborne dust particle prediction.