| Summary: | Environmental sustainability and air quality management are two major issues with regard to air pollution in Malaysia. Many approaches have been introduced to deal with the issues and are yet to be improved. This study aims to examine the effect of Using Principal Component Analysis (PCA), Discriminant Analysis (DA), and Artificial Neural Network (ANN), in finding the most significant air pollution and to propose a model that integrates both PCA, DA, and ANN to predict air pollution. A web based enviroment for air pollution index was developed in order to display the prediction result of air pollution in a few selected regions in Malaysia. These regions are: Johor, Terengganu, Pinang, Sarawak and Selangor. Multivariate statistical technique was applied to model the pattern and dynamic characteristics of air pollutants. PCA was adopted to analyse the interrelationship among the pollution variables. Its operation is based on common underlying dimension used to reduce the dimensionality of the complex data set. It is used to identify the major possible sources of pollution by providing empirical estimates of the variables. DA is one of the data mining techniques used to discriminate a single classification variable using multiple attributes. It is used to identify the most important parameters that best discriminate in a given data set. ANN was used to learn the complex data patterns and applied to the activities of prediction. The experimental results showed that PCA gave a strong variance of over 30% and 21% for Varimax Factor (VF) 1 and 2 respectively. This indicates that the main sources of atmospheric air pollution are from industrial activities, vehicle emission, and power plants. DA gave a correct assignation of over 86%, indicating that only three parameters; Particulate Matter (PM10), Nitrogen Dioxide (NO2) and Sulphur Dioxide (SO2) discriminate best with a P-value < 0.0001. The result from ANN gave a Coefficient of Determination (R2) = 79%, 72% and 78% for training, testing and validation with Root Mean Square Error (RMSE) = 6.61, 5.77,4.61 respectively. This finding indicates that ANN can predict over 70% of the air pollution index in the five selected regions. The application of multivariate techniques, computer based model and ANN revealed basic information that can be used by the government, other stakeholders and the potential researchers in air quality modelling and decision making.
|
|---|