AUTOENCODER ARTIFICIAL NEURAL NETWORK MODEL FOR AIR POLLUTION INDEX PREDICTION

Air pollution, a significant global challenge driven by industrialization, urbanization, and population growth, is caused by the emission of harmful gases, particulates, and biological molecules into the atmosphere, posing serious risks to health and the environment. Key sources include power plants, industrial activities, vehicles, and residential heating. Thus, effective air quality monitoring and forecasting are crucial to mitigating the adverse impacts of pollution. This paper presents shallow and deep sparse autoencoder artificial neural network models to improve the prediction of the Air Pollution Index (API) in Perak Darul Ridzuan, Malaysia, as a case study. The results show that the deep sparse autoencoder achieves better prediction accuracy with MSE and R2 values of 0.1474 and 0.8331, respectively, compared to 0.1515 and 0.8300 for the shallow sparse autoencoder. The performance of these autoencoder models is also compared with other models, such as feedforward artificial neural networks (FANN) and principal component analysis (PCA). The findings confirm that both autoencoder models enhance API prediction accuracy, with the deep sparse autoencoder emerging as the optimal model, highlighting the potential of deep learning in improving air quality prediction.