Handling Generalization in Contextualized Support Vector Machine Knowledge Models
Marcelo Loor, Ana Tapia-Rosero, Guy De Tré · 2025
The contextualization of support vector machine (SVM) knowledge models has recently been proposed as a means to enhance the interpretability of predictions generated by SVM classifiers. Nevertheless, a possible drawback of that approach is the potential loss of generalization in the resulting contextualized models. Aiming to overcome this drawback and further improve interpretability, we propose in this paper a novel approach to generalization of contextualized SVM knowledge models. In our approach, a similarity comparison between two augmented intuitionistic fuzzy sets representing the evaluations respectively performed by two models is used as an indicator of the similarity between these models. The results of such comparisons are then used to cluster similar models. The reduced number of clustered models can be used afterwards for making contextualized predictions. An example in which skin lesions are identified illustrates how the detection and clustering of similar contextualized models can favor their generalization while improving the interpretability of the resulting predictions.