Preprocessing and Optimization of Smooth Data-driven Model for Emergency Conditions Against Air Pollution

摘要

Magnitudesof theairpollutiondepend onvariousvariables.Preprocessing and optimisation processes are necessary todiscover the complexity of the relationship of the data formore accurate and eff icient predictions. These techniqueshelp to clean the datasets and to f ind the best structure ofthe smooth data model. The Gamma test (GT) and GeneticAlgorithm (GA) are practical tools which can be applied forpreprocessing and optimising data models. Regardingbuilding a smooth data model, the developed artif icialneural networks are functional optimisation strategieswhich are suitable for ANN training. Moreover, local linearregression (LLR) and dynamic local linear regression(DLLR) models are effective due to the high density of ournormalised dataset. In thisregard, wedeveloped aprocess toconstruct a smooth data model to support environmentaldecision making in airpollution emergencyconditions. Themain objective of this work was to set an appropriatealgorithm by preprocessing and optimising a set of the datamodel for developing smooth data-driven models whichcould play a signif icant role in early warning systems inregard to the human health. The data sets included themeteorological and air pollutant variables asinputs/predictors and emergency medical service clients asoutputs. The GT and GA were applied to analyse andoptimise the input variables. Three types of ANNS (ANN1,ANN2, and ANN3), (LLR), and (DLLR) techniques wereused to establish the models. Finally, a smooth data modelwasconstructed and evaluated.