IoT-based Electronic Nose and Environmental Sensor Fusion for Melon Root Rot Classification Using Adaptive Feature Optimization

This study proposes an IoT-based sensor fusion framework for melon root rot detection and severity classification by integrating a six-sensor electronic nose array with soil pH, electrical conductivity, and humidity sensors into a unified nine-dimensional feature space.The principal contribution is a classifier-adaptive sequential forward selection instantiation, wherein each of eight classifiers spanning classical machine learning, neural network, and deep learning paradigms independently identifies its optimal feature subset from the fused space.Evaluated across six experimental scenarios with stratified ten-fold cross-validation, hold-out testing, and Wilcoxon signed-rank analysis -with bootstrap resampling and standardization applied exclusively within the training combination of each cross-validation fold after fold assignment, ensuring validation folds remain untouched at all stages -all eight models achieved perfect classification accuracy of 1.0000 on the locked hold-out test set using only three to eight features, with cross-validation accuracy under the proposed S6 scenario ranging from 0.9865 to 0.9973 across the eight models, while hold-out test accuracy ranged more broadly from 0.6957 to 1.0000 across all six evaluated scenarios depending on the feature configuration, and while delivering training time reductions up to 26.16% and model size reductions up to 28.55% for classical and neural network architectures under controlled laboratory conditions.Soil pH and humidity emerged as universal cross-architecture biomarkers, providing evidence-based guidance for minimal-sensor IoT node design in precision agriculture.