Machine Learning Techniques to Predict the Air Quality Using Meteorological Data in Two Urban Areas in Sri Lanka

Authors

Lakindu Mampitiya, Water Resources Management and Soft Computing Research Laboratory
Namal Rathnayake, The University of Tokyo
Lee P. Leon, University of the West Indies
Vishwanadham Mandala, Indiana University
Hazi Md. Azamathulla, University of the West Indies
Sherly Shelton, University of Nebraska-Lincoln
Yukinobu Hoshino, Kochi University of Technology
Upaka Rathnayake, Atlantic Technological University

Date of this Version

8-8-2023

Citation

Mampitiya, L.; Rathnayake, N.; Leon, L.P.; Mandala, V.; Azamathulla, H.M.; Shelton, S.; Hoshino, Y.; Rathnayake, U. Machine Learning Techniques to Predict the Air Quality Using Meteorological Data in Two Urban Areas in Sri Lanka. Environments 2023, 10, 141. https://doi.org/10.3390/ environments10080141

Comments

Open access.

Abstract

The effect of bad air quality on human health is a well-known risk. Annual health costs have significantly been increased in many countries due to adverse air quality. Therefore, forecasting air quality-measuring parameters in highly impacted areas is essential to enhance the quality of life. Though this forecasting is usual in many countries, Sri Lanka is far behind the state-of-the-art. The country has increasingly reported adverse air quality levels with ongoing industrialization in urban areas. Therefore, this research study, for the first time, mainly focuses on forecasting the PM₁₀ values of the air quality for the two urbanized areas of Sri Lanka, Battaramulla (an urban area in Colombo), and Kandy. Twelve air quality parameters were used with five models, including extreme gradient boosting (XGBoost), CatBoost, light gradient-boosting machine (LightBGM), long short-term memory (LSTM), and gated recurrent unit (GRU) to forecast the PM₁₀ levels. Several performance indices, including the coefficient of determination (R²), root mean squared error (RMSE), mean absolute error (MAE), mean squared error (MSE), mean absolute relative error (MARE), and the Nash–Sutcliffe efficiency (NSE), were used to test the forecasting models. It was identified that the LightBGM algorithm performed better in forecasting PM₁₀ in Kandy (R² = 0.99, MSE = 0.02, MAE = 0.002, RMSE = 0.1225, MARE = 1.0, and NSE = 0.99) . In contrast, the LightBGM achieved a higher performance (R² = 0.99, MSE = 0.002, MAE = 0.012, RMSE = 1.051, MARE = 0.00, and NSE = 0.99) for the forecasting PM10 for the Battaramulla region. As per the results, it can be concluded that there is a necessity to develop forecasting models for different land areas. Moreover, it was concluded that the PM₁₀ in Kandy and Battaramulla increased slightly with existing seasonal changes.