PEMODELAN DERET WAKTU KONSENTRASI PM10 DI KOTA SURABAYA DAN HUBUNGANNYA DENGAN FAKTOR METEOROLOGI DENGAN METODE EXPONENTIAL SMOOTHING DAN MACHINE LEARNING

Abstract

Air pollution remains a major environmental issue in urban areas, particularly in Surabaya, where particulate matter with an aerodynamic diameter of less than 10 µm (PM₁₀) poses significant risks to human health and the environment. This study aims to model the time series of PM₁₀ concentrations in Surabaya and to examine their relationship with meteorological variables, including average temperature (TAVG), relative humidity (RH), wind speed (FF_AVG), and rainfall (RR). Several statistical and machine learning approaches were employed, namely Exponential Smoothing, Multiple Linear Regression (MLR), Gradient Boosting Regression (GBR), and Random Forest (RF). Model performance was evaluated using Root Mean Square Error (RMSE), Mean Absolute Error (MAE), Mean Absolute Percentage Error (MAPE), and the Nash–Sutcliffe Efficiency (NSE), followed by an aggregation ranking analysis. The results show that Multiple Linear Regression achieved the best overall performance, particularly on the Kebonsari-exclusion dataset, with an RMSE of 18.01 µg/m³, MAE of 13.41 µg/m³, MAPE of 28.10%, and an NSE of 0.034, indicating more stable and accurate predictions compared to the other methods. Forecasting results using the MLR model across four dataset variants—Wonorejo-imputation, Kebonsari-imputation, Wonorejo-exclusion, and Kebonsari-exclusion—over a three-day horizon demonstrated consistent trends and robust performance, regardless of whether missing data were handled through imputation or exclusion. Furthermore, analysis of the relationship between PM₁₀ and meteorological factors revealed spatial variability in dominant predictors. At SPKUA Wonorejo, average temperature (TAVG) was identified as the most influential variable, showing a positive relationship with PM₁₀ concentrations. In contrast, wind speed (FF_AVG) emerged as the dominant factor at SPKUA Kebonsari, where higher wind speeds were associated with lower PM₁₀ concentrations. Overall, this study confirms the effectiveness of Multiple Linear Regression for PM₁₀ forecasting and provides methodological insights to support air quality management in Surabaya.