Forecasting of PM2.5 in Malaysia using hybrid artificial neural network / Pavithra Chinatamby

Pavithra , Chinatamby (2023) Forecasting of PM2.5 in Malaysia using hybrid artificial neural network / Pavithra Chinatamby. Masters thesis, Universiti Malaya.

	PDF (The Candidate's Agreement) Restricted to Repository staff only Download (164Kb)
	PDF (Thesis M.A.) Download (3739Kb)

Abstract

Particulate matter with aerodynamic diameter of less than 2.5 microns (PM2.5) is becoming a prominent air pollutant in our atmosphere currently and it causes serious effects to human health upon prolonged exposure. Forecasting the level of PM2.5 earlier can help us to take necessary actions to avoid the exposures but identifying the best prediction model that can handle the big air quality data is a huge challenge. This research is aimed to identify the most reliable prediction model to predict the level of PM2.5 pollutant. The air quality and meteorological data measured at industrial areas from the year 2017 to 2019 were collected from Department of Environment (DOE), Malaysia. The prediction model was build using multi-layered feedforward artificial neural network (FANN) which is a type of artificial neural network (ANN) method. FANN model with six different training algorithms with thirteen different training functions were analysed, compared and top five different training functions were selected. Then, hybrid models were created where FANN model was infused with dimensionally reduced data using Principal Component Analysis (PCA) and Partial Least-Squares (PLS) techniques respectively. The performance of hybrid models such as PCA-FANN and PLS-FANN were compared with the FANN model and other regression models. The regression models analysed in this research are multiple linear regression (MLR), principal component regression (PCR) and partial least squares regression (PLSR). All these models were evaluated based on the highest R2 value and lowest RMSE, MAE and MAPE values obtained. The ascending order according to the modelling performance are PCR< MLR< PLSR< PCA-FANN< PLS-FANN. The PLS-FANN hybrid model with Levenberg Marquardt (trainlm) as training function has topped the ranking by obtaining the highest R2 value of 0.9999 with the lowest RMSE and MAE values for the testing datasets which are 0.001 and 0.0005 respectively. It is the best performed and the most reliable model compared to the other models.

Actions (For repository staff only : Login required)