Rahim Nur Alis Addiena A, Noor Norazian Mohamed, Jafri Izzati Amani Mohd, Ul-Saufie Ahmad Zia, Kamaruddin Mohamad Anuar, Zainol Mohd Remy Rozainy Mohd Arif, Sandu Andrei Victor, Vizureanu Petrica, Deak Gyorgy
Faculty of Civil Engineering Technology Universiti Malaysia Perlis Arau Perlis Malaysia.
Sustainable Environment Research Group (SERG) Centre of Excellence Geopolymer and Technology (CEGeoGTech) Arau Perlis Malaysia.
Anal Sci Adv. 2025 Jul 15;6(2):e70027. doi: 10.1002/ansa.70027. eCollection 2025 Dec.
Severe haze episodes in Southeast Asia, largely attributed to transboundary pollution from neighbouring countries, have led to substantial environmental degradation and adverse health effects. This study presents the development of a novel air quality forecasting model specifically developed to predict particulate matter with a diameter of less than 10 µm (PM) concentrations 1 to 3 days in advance during transboundary haze events in Malaysia. The innovation lies in the integration of quantile regression (QR) with advanced feature selection techniques-namely, Relief-based ranking, correlation-based selection and principal component analysis (PCA)-to form modified predictive models. These hybrid models, referred to as QR-Relief, QR-correlation and QR-PCA, demonstrated superior performance over traditional QR and multiple linear regression models across four urban locations: Klang, Melaka, Pasir Gudang and Petaling Jaya. Model accuracy was evaluated using selected performance metrics, including mean absolute error, normalized absolute error and root mean square error. The results indicate that reducing the dimensionality of input variables through analytical feature selection significantly improves predictive reliability. Furthermore, model validation using an independent dataset from 2019 confirmed their real-world applicability. This methodological advancement provides a robust analytical framework for developing early warning systems during haze events, offering valuable decision-support tools for environmental and public health management.
东南亚严重的雾霾事件主要归因于邻国的跨境污染，已导致严重的环境退化和对健康的不利影响。本研究介绍了一种新型空气质量预测模型的开发，该模型专门用于在马来西亚跨境雾霾事件期间提前1至3天预测直径小于10微米的颗粒物（PM）浓度。创新之处在于将分位数回归（QR）与先进的特征选择技术（即基于Relief的排序、基于相关性的选择和主成分分析（PCA））相结合，形成改进的预测模型。这些混合模型，即QR-Relief、QR-相关性和QR-PCA，在巴生、马六甲、巴西古当和八打灵再也这四个城市地点，表现优于传统的QR和多元线性回归模型。使用选定的性能指标（包括平均绝对误差、归一化绝对误差和均方根误差）评估模型准确性。结果表明，通过分析特征选择降低输入变量的维度可显著提高预测可靠性。此外，使用2019年的独立数据集进行的模型验证证实了它们在现实世界中的适用性。这一方法的进步为雾霾事件期间开发预警系统提供了一个强大的分析框架，为环境和公共卫生管理提供了有价值的决策支持工具。
