Data driven particulate matters prediction in Klang Valley using machine learning techniques / Neo En Xin

Neo , En Xin (2024) Data driven particulate matters prediction in Klang Valley using machine learning techniques / Neo En Xin. PhD thesis, Universiti Malaya.

	PDF (The Candidate's Agreement) Restricted to Repository staff only Download (342Kb)
	PDF (Thesis PhD) Download (4Mb)

Abstract

The environment has been significantly impacted by rapid urbanization, leading to a need for changes in climate change and pollution indicators. An air quality intelligence platform operates as a complete measurement site for monitoring and governing air quality, shown promising results in providing actionable insights. This approach offers a holistic strategy to tackle air quality issues and their impact on climate change and human health. This research aims to highlight the potential of machine learning models in forecasting air quality and providing data-driven strategic and sustainable solutions for smart management. An end-to-end air quality predictive model is proposed utilizing four machine learning techniques and three deep learning techniques namely: 1) Support Vector Regressor (SVR), 2) random forest (RF), 3) Gradient Boosting (GB), 4) k-nearest neighbour (KNN), 5) long short-term memory (LSTM), 6) Bi-directional LSTM (Bi-LSTM) and 7) stacked LSTM. In the proposed work, four different urban cities (cities of Petaling Jaya, Banting, Klang and Shah Alam) in Selangor are considered and the air quality data from the year 2010 to 2016 and 2018 to 2019 in the regions are collected from the Department of Environment (DOE), Malaysia. In addition, medical data such as lung diseases due to external agent (ICD-10 J60-J70) are collected from Ministry of Health (MOH) for the prediction of hospital admission count. The model considered the air quality data of various pollutant markers such as particulate matter (PM2.5 and PM10), ozone (O3), carbon monoxide (CO), nitrogen dioxide (NO2), sulfur dioxide (SO2), and meteorological data such as wind speed, wind direction, temperature, and humidity. The results of the forecasting of the concentration of the PM2.5 and PM10 are presented in R2, RMSE and MAE. The results also showed that the involvement meteorological parameters in the environmental data improves the forecasting performance. From the forecasting of PM10 and PM2.5 concentration, Bi- LSTM performed overall the best results with R2 and RMSE values of 0.9998,0.0031; 0.9978, 0.0053; 0.9989, 0.0042 and 0.9968, 0.0082 in Dataset A and 0.8896, 0.0387; 0.8882, 0.04; 0.9109, 0.0400, and 0.7063, 0.0163 in Dataset B for PM10. While the results for PM2.5 are 0.8880, 0.0400; 0.9057, 0.0470, 0.9147, 0.0387 and 0.6794, 0.0155 in Banting, Petaling, Shah Alam and Klang stations, respectively. Besides, feature optimization is proposed to reduce dimensionality and remove irrelevant features to enhance the estimation of the concentration of the pollutants in this study. It indicated PM10, PM2.5, CO, NO2, SO2, wind direction, humidity and temperature are the most important elements to monitor PM10 and PM2.5. By reducing the number of features used in the model (i.e. only 6 important features), the proposed feature optimization process enables the model to be more interpretable and provides insights into the most critical factor affecting air quality. Furthermore, the optimized model is further implemented to predict the hospital admission counts due to air pollution. Findings from this study can aid policymakers to understand the underlying causes of air pollution and develop more effective smart strategies for reducing pollution levels.

Actions (For repository staff only : Login required)