Domain-driven KDD for mining functionally novel rules and linking disjoint medical hypotheses
Knowledge-Based Systems
| Hi-index | 0.00 |
We introduce a novel domain-driven rule discovery and evaluation algorithm based on Swanson’s logical relation approach. Over more than a decade, rules have been mined from large biomedical datasets and been evaluated solely based on statistical properties of the rules or user-belief specifications. This approach faces tremendous challenges to determine novel, actionable and interesting rules. In this paper, we introduce a new paradigm in addressing rule interestingness problem using domain knowledge. We demonstrate that novel and interesting association rules can be discovered from large medical datasets based on its ability to infer previously unknown relations in biomedical domain. Our data mining algorithm shows that we can effectively achieve this task by incorporating biomedical domain knowledge by combining both literatures and ontology. We outline the conceptual-architectural framework for future implementation of this methodology.