View publication
| Title | An Empirical Analysis of Feature Engineering for Dempster-Shafer Classifier as a Rule Validator |
| Authors | Aneta Baloyan, Alexander Aramyan, Nelson Baloian, Arnak Poghosyan, Ashot Harutyunyan, Sergio Peñafiel |
| Publication date | 2024 |
| Abstract | Explainable AI methods are increasingly attracting the practitioner's attention since they convey important information about the nature of the phenomena being studied. Rule-generating models are considered one of the most explainable ones, however, there are so far not many attempts aimed at evaluating the rules generated by this kind of models. This paper proposes using an explainable classifier based on the Dempster-Shafer (DS) plausibility theory as a rule-validating mechanism to assess the reliability of the rule sets generated by AI models. The DS theory enables combining evidence from various sources while dealing with conflicting or even contradictory information, identifying trustworthy rules with high belief correctness. Our empirical analysis evaluates the DS-based classifier's ability to learn possibly complex numeric feature interactions on synthetic datasets. Results show the model excels at numeric interactions like differences and ratios and performs well regardless of class imbalance, but struggles with imbalanced categorical data. We conclude by proposing future work including comparative result analysis against traditional methods and developing hybrid approaches for more robust but at the same time interpretable AI systems. |
| Pages | 143-152 |
| Conference name | Online Workshop on Collaborative Technologies and Data Science in Smart City Applications |
| Publisher | Logos Verlag (Berlin, Germany) |
| Reference URL |
|