Using mineral profiles to build models to classify the geographic origin of applies is complicated by substantial variability within and between production regions.
This study (open access) examined the impact of geographical class definition and dataset structure on the classification of apple origin using linear discriminant analysis (LDA), and whether these effects could support a stepwise classification strategy.
The authors propose a hierarchical multiple-model classification strategy as an alternative to relying on a single global model. In such a system, a broad model could first identify well-separated geographical regions and potential collision regions. (a collision region was defined as a geographical class showing cross-class misclassification together with overlap with at least one other class in the canonical LDA space), Then, dedicated models could focus on discrimination among closely related regions
As a case-study, the mineral and isotopic profiles of four apple cultivars (‘Gala’, ‘Golden Delicious’, ‘Jonagold’, and ‘Idared’) originating from the Czech Republic and Poland were evaluated using five LDA models that differed in terms of geographical class definition, dataset size, and macro- versus micro-regional grouping.
The researchers found that the geographical classification of apple samples based on mineral and isotopic profiles was significantly impacted by the definition and structure of geographical regions. Decomposing macro-regions into micro-regions did not universally improve classification. However, it revealed substantial intra-regional heterogeneity and allowed for the identification of collision regions. Improved performance observed after excluding previously identified collision regions further indicated that a substantial part of the classification uncertainty was concentrated in a limited number of geographical regions rather than being uniformly distributed across the dataset.
The authors consider that these findings support further investigation of a hierarchical multiple-model classification strategy as an alternative to relying on a single global model. In such a system, a broad model could first identify well-separated geographical regions and potential collision regions. Then, dedicated models could focus on discrimination among closely related regions.
Developing such an approach would require representative micro-regional reference datasets, careful variable selection with sufficient geographical discriminatory power, consideration of cultivar and agronomic effects, and independent external validation. The proposed framework may therefore represent a step toward more flexible and interpretable authentication of geographical origin based on mineral profiles.
The authors plan to validate this approach, for apples, on independent datasets spanning additional years and a wider range of agricultural crops to assess its broader applicability.
Comments