DistTDT: distributed terminological decision tree learning

The induction of terminological decision trees (TDTs) is among the most widely used algorithms for classifying data and discovering implicit rule-based knowledge from data, which greatly supports decision-making. Centralized models of these algorithms fail to process large-scale datasets due to the limitations of both software and hardware infrastructure. This limitation motivated us to develop the Distributed Terminological Decision Tree (DistTDT), a decentralized approach for inducing terminological decision trees that significantly improves the results of concept learning problems. We integrate a refinement operator into DistTDT to generate a set of subsumed concept candidates. Furthermore, we introduce a prototype for a distributed reasoner, which is necessary for class satisfiability checks, retrieving instances of a specified class, and performing entailment checks. Empirical evaluation demonstrates that the proposed model outperforms existing baselines by implementing enhancements to existing learning algorithms. DistTDT enables efficient processing of large-scale Web Ontology Language (OWL) datasets and scales horizontally. Finally, our approach outperforms a well-known centralized approach in terms of runtime performance and learning metrics, including matching rate, commission error rate, omission error rate, and induction rate. DistTDT is publicly available as an open-source extension integrated into the SANSA framework.