Merged Methods of Artificial Neural Networks and Statistical Techniques in Forecasting Air Quality in the Northern Region of Peninsular Malaysia

Abstract Various disciplines employed artificial intelligence (AI) as a tool in research studies. As part of AI, an artificial neural network (ANN) is notably recognized as a predominant computational tool in air quality studies due to its capabilities in the prediction of gaseous and particulate pollutant concentration as well as forecasting air pollutant index (API). This research was undertaken to discover the potential contributor of air pollutants, the most significant air pollutant in air quality variations, and investigate the forecasting capabilities using the merged methods known as the ANN-Sensitivity Analysis Model (ANN-SAM) and ANN-Principal Component Analysis (ANN-PCA). Initially, ANN-SAM discovered that only four out of six are considered major pollutants which are ozone (O₃), sulfur dioxide (SO₂), particulate matter (PM₁₀), and particulate matter (PM2.5). While ANN-PCA discovered that all are major pollutants, except sulfur dioxide (SO₂). Based on the initial findings, further analysis was conducted by employing new merged models which are known as MLP-FF-ANN-SAM and MLP-FF-ANN-PCA. The MLP-FF-ANN-SAM model used four pollutants as input parameters, while the MLP-FF-ANN-PCA model used five pollutants. The results for both models showed a strong correlation with R² values of 0.826 and 0.821, and RMSE values of 5.982 and 5.922, accordingly. This indicates that both merged models are suitable for forecasting air quality. Besides, these models are also capable of forecasting air quality using fewer parameters with reliable outcomes.