IoT based smart framework to predict air quality in congested traffic areas using SV-CNN ensemble and KNN imputation model
Khaled Alnowaiser,Aisha Ahmed Alarfaj,Ebtisam Abdullah Alabdulqader,Muhammad Umer,Lucia Cascone,Bhavya Alankar
DOI: https://doi.org/10.1016/j.compeleceng.2024.109311
IF: 4.152
2024-06-01
Computers & Electrical Engineering
Abstract:Addressing air pollution presents a significant environmental challenge in the context of smart city environments. Real-time monitoring of pollution data empowers local authorities to assess current traffic conditions and make informed decisions accordingly. The integration of Internet of Things (IoT) sensors has revolutionized air quality prediction, with human activities being the primary contributors to air pollution, posing threats to all forms of life. Gases such as SO2, PM10, O3, CO, NO2, among others, are key pollutants. Exposure to air pollution can lead to severe health issues and fatalities, underscoring the criticality of air quality monitoring. Distinguishing between breathable and non-breathable air quality further enhances the value of air quality monitoring. However, existing techniques face challenges in achieving high accuracy, exacerbated by missing values in datasets, which can significantly impact machine learning model performance. In response to these challenges, this research introduces an IoT-based automated system designed to classify air quality. This system is adept at managing missing data and attaining high accuracy levels. Central to this approach is a stacked ensemble voting classifier model, which combines two machine learning models. Additionally, the system incorporates the KNN Imputer to address missing values effectively. The work assesses the system's performance against seven alternative machine learning algorithms across two scenarios: one with missing values removed and another with KNN imputation applied. Notably, the proposed strategy achieves remarkable metrics, including 99.17% accuracy, 97.75% precision, 95.24% recall, and a 96.52% F1 score, when leveraging the KNN Imputer. These results underscore the effectiveness of the proposed model compared to current state-of-the-art methodologies.
engineering, electrical & electronic,computer science, interdisciplinary applications, hardware & architecture