Search

This presentation will discuss the used of an artificial intelligent method namely the ‘stochastic boosted regression trees’ (BRT) approach that uses an algorithm that applied to an air pollution data namely particle number count concentrations ([PNC]), an ultrafine particles data and particulate matter data case study in United Kingdom and Malaysia. The development of the BRT model involves determining the model algorithm settings of the main model input parameters (learning rate, number of trees and interaction depth) that were tested using the R software (version 3.02) by choosing a10-fold cross-validation approach with combination of lr 0.05 and tc 5 of training set for BRT models. It was found, that the coefficient of determination (R2) value for the BRT best iteration models were above 0.60 for [PNC] in urban environment. The fine and course particle number (FPNC and CPNC) were found to be 0.75 and 0.72 respectively for one of coastal dataset while R2 value of 0.78 and 0.85 were obtained for Malaysia data. Further investigated were performed to rank factor influenced. It was found, that Carbon monoxide (30.28 %) gas and followed by temperature (16.81%) and wind direction (16.4%) were found the high factor influenced PM10 in urban environment. The interaction index (H-index) between parameters to concentration of pollutants were also examined graphically and in numerical form (H-Index). It was found that the H-Index between parameters 0.3 to 0.4 indicated that the BRT technique able to explain the science of air pollution. The consistent results to produce the best model from the best iteration, able to rank the best parameters that influence most to the concentration of predictor and able to predict interaction between variables premise BRT as one of the method or tools to analyse air pollution data.