Trend and missing data prediction model of PM10 in central region using ANN and MLR

Nurul Latiffah, Abd Rani and Azman, Azid (2018) Trend and missing data prediction model of PM10 in central region using ANN and MLR. In: 31st International Conference of Analytical Sciences (SKAM31), 17 Aug 2018, Kuantan, Pahang.

Text FH03-FBIM-18-18278.pdf Restricted to Registered users only Download (89kB)

Abstract

Increasing concentrations of PM10 are identified to give harmful affect to human health. Trend analysis of PM10 for the last six years starting from 2010 until 2015 in Central region Malaysia shows increased and decrease concentration of the PM10 pollutant. Some of the PM10 concentration exceeds the Malaysia Ambient Air Quality Guidelines which is 150 µg/m3 . In addition, the high Air Pollution Index (API) was assign to the high PM10 concentrations which was the main pollutant in the air. Despite their importance in determining the API level in Malaysia, there are some missingness data of PM10 detected for certain day perhaps due to the failure of the equipment. Therefore, missing data prediction of PM10 may give vital information with the intention of taking actions for the public and government especially regarding the API levels which is the main indicator used to decide the level of air quality. There are eight continuous air quality monitoring station in Central region which located in Selangor (Klang, Petaling Jaya, Banting, Shah Alam, Kuala Selangor) and Kuala Lumpur (Batu Muda, Putrajaya, Cheras). Meteorological and pollutant parameters analyzed for the missing data prediction model of PM10 in this study include wind speed, wind direction, temperature, humidity and NOx, NO, SO2, NO2, CO, O3 respectively. In this study, Artificial Neural Network (ANN) and Multiple Linear Regression (MLR) models conjointly PCA were used to predict the concentration of PM10 missing data in Central region Malaysia. The results obtained from trend analysis signified that each continuous air quality monitoring station in Central region give different concentrations of PM10 with Klang continuous air quality monitoring station shows the highest concentration which is 581 µg/m3 in June 2013. This possibly owing to the transboundary pollution from the great land and forest fires in Sumatra and Kalimantan, Indonesia especially during the southwest monsoon (May until September) which is contributed to the worsen air quality in Malaysia. Furthermore, locality of Klang continuous air quality monitoring station and activities done within this region also give high PM10 concentration. For the missing data prediction model, inputs to the models obtained from the Principal Component Analysis (PCA) include pearson coefficient with moderate correlation (0.5-0.75) and pearson coefficient with high correlation (>0.75). All parameters (wind speed, wind direction, temperature, humidity, NOx, NO, SO2, NO2, CO, O3) also being used as inputs besides inputs obtained from PCA. Input parameters obtained from pearson coefficient with moderate correlation (0.5 - 0.75) and high correlation (>0.75) seems not suitable to be applied onto Central regions whether by ANN or MLR model. The results showed that all parameters as inputs use for ANN appeared to be promising with R2 up to 0.5343 and RMSE up to 23.95. However, results obtained from MLR analysis using the same input parameters shows less accurate than ANN with R2 and RMSE value obtained are 0.3478 and 27.65 respectively. It is concluded that ANN is capable to predict the missing data concentration of PM10 rather than MLR model.

Actions (login required)

View Item