IoT integrated quantile principal component analysis based framework for toxic pesticides recognition and classification

Abstract

Pesticides present significant concerns regarding environmental sustainability and global stability. This study investigates the types, benefits, and environmental challenges associated with pesticide use. To address these concerns, we developed an innovative Internet of Things (IoT) integrated quantile principal component analysis (QPCA) framework for the recognition of toxic pesticides in smart farming, termed IoT-TPR. The proposed IoT-TPR system is an intelligent electronic nose based on a tin-oxide sensor array, consisting of eight commercial metal–oxide–semiconductor gas sensors, which detect toxic pesticides and transmit the data to the Amazon Web Services cloud for further analysis. A two-stage QPCA preprocessing technique is employed to analyze sensor responses. Subsequently, four classifiers such as radial basis function (RBF), extreme learning machine (ELM), decision tree (DT), and k-nearest neighbor (KNN) are used for comparative performance evaluation. The results indicate that QPCA-KNN achieves the highest accuracy at 99.05%, outperforming other methods across all performance metrics and demonstrating superior classification capability. RBF (96.24%) and ELM (95.81%) also exhibit strong performance, though slightly lower than QPCA-KNN, while DT (92.35%) shows the lowest accuracy but still maintains reasonable performance. Overall, QPCA-KNN emerges as the most effective and robust classification model in this study.