ConstrainTree: Enhancing Decision Trees With Semantic Constraint Validation

Abstract

A growing number of knowledge graphs are being released by diverse sources. The massive volume of linked data aims to provide entities with semantic context. The entities can be validated in relation to their context using the Shapes Constraint Language (SHACL). On the other hand, the increasing usage of Artificial Intelligence (AI) models in productive systems entails a tremendous deal of responsibility in a variety of domains. Predictive models such as linear, logistic regression, and tree-based models are still widely employed because they have a basic structure that allows for explanation. Explanation of models, however, requires determining if the model generated predictions based on human constraints or scientific facts. This work proposes using the semantic context of entities in knowledge graphs to assess prediction models against user-defined constraints, thus providing a theoretical framework for a model-agnostic validation engine based on SHACL. In a subsequent phase, the model validation results are summarized and model-coherently visualized in the case of a decision tree. The performance of the proposed framework is studied in an empirical evaluation suggesting its efficiency.