Abstract
This study investigates the effects of species, geographical origin, and processing on the proximate composition of frog leg meat, with a focus on developing predictive models for processing status. Data were systematically compiled from 18 published studies, yielding 32 entries across 10 edible frog species and multiple processing methods. Proximate composition parameters (moisture, protein, fat, ash) were compared between processed and unprocessed samples, and classification models were trained using moisture content as the primary predictor. Logistic regression and several machine learning algorithms, including Stochastic Gradient Descent, Support Vector Machine, Random Forest, and Decision Tree, were benchmarked under a Leave-One-Study-Out (LOSO) cross-validation framework. Results demonstrated that moisture content alone was sufficient to accurately distinguish processing status, with a critical threshold of ~73% separating processed from unprocessed frog legs. Logistic regression achieved perfect specificity and precision (100%) with an overall accuracy of 96.8%, while other classifiers also performed strongly (>90% accuracy). These findings confirm moisture as a species- and origin-independent marker of processing, offering a simple, rapid, and cost-effective tool for authenticity verification and quality control in frog meat and potentially other niche protein products. Future work should expand sample coverage, validate thresholds across processing types, and integrate biochemical and sensory quality assessments.