ターゲットを絞った元素プロファイリングに基づく多機能分類戦略を用いた小麦粉トルティーヤの真正性検証

Wheat flour tortillas are a staple food in Mexico, yet reliable authentication methods are limited. This study analyzed 20-element profiles in 145 tortilla samples to determine their geographic origin and manufacturing process using multiple classification strategies. Data were preprocessed with autoscaling, mean centering, and their combinations, and then evaluated through principal component analysis, linear discriminant analysis, k-nearest neighbors, and decision tree models. Geographic origin was the main factor affecting elemental composition, while the manufacturing process had a smaller influence. Classification was conducted using binary, multiclass, and multifunction approaches. Decision tree models achieved 100 % sensitivity, specificity, and accuracy across all processing methods for multifunction classification. In contrast, k-nearest neighbors and linear discriminant analysis achieved accuracies of 88 % to 100 % and 83 % to 92 %, respectively. These findings provide a robust, data-driven framework for tortilla authentication, supporting quality control, regulatory oversight, and consumer confidence.