A Robust IQR-Based Pre-processing and Tri-Linear Neural Network Framework for Multivariate Air Pollution Analysis

Escalating global urbanization, industrialization, and motorization underscores the pressing relevance of analyzing air quality from multiple dimensions to address the myriad impacts of these phenomena on the environment and human well-being. This study focuses on several critical issues that hinder the analysis of air pollution, including the absence of values, presence of value outliers, and poorly defined relationships among pollutant, weather, and time variables to improve the reliability and precision of air quality evaluations. To alleviate these challenges, we suggest an advanced model for pre-processing and feature selection. The proposed approach consists of the Imputer – Interquartile range (IQR) – pre-processing technique that resolves issues of outliers and missing data

on the basis of median-centered statistics and a Tri-Linear Fully Connected (TFC) feature selection method that identifies key variables through learning complex interactions among multiple variables. We demonstrate the effectiveness of our approach through experiments conducted on air quality data from multiple regions. Our pre-processing model achieved an outstandingly low Root Mean Square Error (RMSE) of 6.72, indicating improved data stability and predictive accuracy over the previous state-of-the-art methods. In addition, our feature selection method further enhances the overall classification performance, making it possible to conduct air pollution studies more reliably and accurately.

Message: Hi FE, I need help with “A Robust IQR-Based Pre-processing and Tri-Linear Neural Network Framework for Multivariate Air Pollution Analysis” in “Environmental Engineering”