Classification of Fermented Seasoning Using Electronic Nose and Tree-Based Pipeline Optimization Tool (TPOT)

An electronic nose (e-nose) system for classifying four Filipino fermented seasonings: Bagoong Alamang, Patis, Suka, and Achara was developed in this study. Utilizing a cost-effective Orange Pi One platform and a sensor array comprising MQ-2, MQ-3, MQ-4, MQ-5, MQ-6, MQ-8, and MQ-135 gas sensors, we address the challenge of limited sample availability commonly found in e-nose applications. Our system leveraged the Tree-based Pipeline Optimization Tool (TPOT), an AutoML approach, to identify an optimized pipeline from 158 training samples with 40 validation samples across 8 different products from varying brands. The resulting exported pipeline from TPOT uses the K-Nearest Neighbors algorithm, which achieved ${1 0 0 \%}$ classification accuracy on the testing set with a TPOT cross-validation (CV) score of 97.85%. While we achieved perfect classification accuracy, we acknowledge this may present an overly optimistic assessment due to training on single samples per brand, potentially limiting real-world generalizability. Time-series data augmentation significantly enhanced the dataset, successfully addressing sample availability limitations.