A Hybrid Decision Logic Framework Combining Fuzzy Systems and Machine Learning Classifier for Post-Fermentation Tea Quality Grading

This study addresses the need for reliable post-fermentation tea quality grading by proposing a hybrid decision logic framework that com-‎bines narrowband spectrophotometric sensing, computer vision-based fermentation indicators, and intelligent classification techniques. The ‎proposed methodology employs low-cost sensing and digital imaging for feature extraction, followed by a two-stage modeling approach ‎that integrates rule-based fuzzy inference system (FIS) with machine learning (ML) classification. After feature extraction, statistical and ‎ML-based feature selection techniques were applied to identify the most informative features. The initial stage employs a FIS to identify the ‎grade and its confidence score based on recurrent fuzzy patterns corresponding to each tea grade. In the second stage, a light-gradient boost‎ing machine LightGBM classifier is trained to enhance predictive accuracy and generate probabilistic confidence scores. Finally, hybrid ‎decision logic is applied to finalize the grade. The proposed hybrid model achieved an accuracy of 99.00%, precision of 99.53%, recall of ‎‎98.27%, and F1-score of 98.74%, outperforming both traditional ML and state-of-the-art models. It demonstrated robust generalization ‎across varying sample sizes and cross-validation folds. Ablation studies confirmed the efficiency of the FIS, showing strong performance ‎with reduced feature sets. Furthermore, the framework maintained shorter execution times than ensemble-based methods, making it suitable ‎for real-time and resource-constrained environments‎.