Scholarlink

<p>Global health is threatened by air pollution, as particulate matter (PM<sub>10</sub> and PM<sub>2.5</sub>) exposure worsens cardiopulmonary diseases. Malaysia’s haze conditions mandate machine-learning forecasts beyond traditional linear models. This study examines Support Vector Regression (SVR) and Multiple Linear Regression (MLR) using five years of PM<sub>10</sub> and PM<sub>2.5</sub> monthly averages data (2018–2022) from four regulatory monitoring stations in Pulau Pinang to assess the efficacy of SVR and MLR models in forecasting particulate matter concentrations, predictive accuracy, error characteristics, and robustness in haze events. The analysis incorporates atmospheric and meteorological predictors; carbon monoxide (CO), nitrogen dioxide (NO₂), ozone (O₃), Sulphur dioxide (SO₂), temperature, and aerosol optical depth (AOD). SVR consistently outperformed MLR across all locations, achieving higher coefficient of determination (R² = 0.88–0.99), lower root mean square error (RMSE = 2.1–0.71), and reduced mean absolute error (MAE = 0.96 − 0.28) relative to MLR. A scatter index (SI = 0.080 to 0.040) supports relative performance improvements by enhancing error homogeneity and stability. SVR’s performance index (PI) consistently exceeded MLR’s between 0.66 and 0.89, indicating improved generalization and dependability suggesting that SVR offers better generalization and dependability in its performance. SVR predicted the peak size and onset date of southwest monsoon haze with 50% to 65% accuracy. Seasonal investigations revealed that regional smoke transport and climatic effects increased PM<sub>2.5</sub> concentrations during haze events. These findings indicate that SVR is an effective framework for forecasting particulate matter in monsoon-affected urban areas, showing its potential for integration into Malaysia’s real-time air quality warning and haze management systems.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Comparative analysis of machine learning models for predicting PM₁₀ and PM2.5 Concentrations in Pulau Pinang, Malaysia

  • 期刊论文

  • 发表日期：

    2026年2月2日

查看全文
  • Emmanuel Yohanna,
  • Lim Hwee San

摘要

Global health is threatened by air pollution, as particulate matter (PM10 and PM2.5) exposure worsens cardiopulmonary diseases. Malaysia’s haze conditions mandate machine-learning forecasts beyond traditional linear models. This study examines Support Vector Regression (SVR) and Multiple Linear Regression (MLR) using five years of PM10 and PM2.5 monthly averages data (2018–2022) from four regulatory monitoring stations in Pulau Pinang to assess the efficacy of SVR and MLR models in forecasting particulate matter concentrations, predictive accuracy, error characteristics, and robustness in haze events. The analysis incorporates atmospheric and meteorological predictors; carbon monoxide (CO), nitrogen dioxide (NO₂), ozone (O₃), Sulphur dioxide (SO₂), temperature, and aerosol optical depth (AOD). SVR consistently outperformed MLR across all locations, achieving higher coefficient of determination (R² = 0.88–0.99), lower root mean square error (RMSE = 2.1–0.71), and reduced mean absolute error (MAE = 0.96 − 0.28) relative to MLR. A scatter index (SI = 0.080 to 0.040) supports relative performance improvements by enhancing error homogeneity and stability. SVR’s performance index (PI) consistently exceeded MLR’s between 0.66 and 0.89, indicating improved generalization and dependability suggesting that SVR offers better generalization and dependability in its performance. SVR predicted the peak size and onset date of southwest monsoon haze with 50% to 65% accuracy. Seasonal investigations revealed that regional smoke transport and climatic effects increased PM2.5 concentrations during haze events. These findings indicate that SVR is an effective framework for forecasting particulate matter in monsoon-affected urban areas, showing its potential for integration into Malaysia’s real-time air quality warning and haze management systems.