A Robust Semantic Web Framework for Actionable Knowledge Discovery (AKD)

Abstract

The growing demand for extracting meaningful and actionable insights from data mining and knowledge discovery systems has led to a paradigm shift from the Knowledge Discovery Process (KDP) to the Actionable Knowledge Discovery (AKD) process. However, significant challenges remain in actionable knowledge discovery. First, the lack of integration between human domain expertise and mining techniques restricts generating practically useful patterns. Second, existing approaches typically rely on a single subjective interestingness measure (e.g., actionability, unexpectedness, or novelty) during post-analysis, thereby limiting their real-world applicability. To address these gaps, this research developed a Semantic Web-based Actionable Knowledge Discovery (SEM-AKD) model, which systematically incorporates domain knowledge throughout the knowledge discovery pipeline. The model organizes the entire actionable knowledge discovery process into three integrated tiers: Data-tier (pre-processing stage), Data Mining tier, and Knowledge Discovery tier (post-analysis stage). This model used the concept of the semantic web (Linked Open Data (LOD)) at the pre-processing stage to improve the accuracy of the classification algorithm by including additional data to the dataset from the linked open data while employing a hybrid filter method for the feature selection. The new model used the Hybridized Multi-Criteria Decision Making (MCDM) model to rank the patterns for their actionability in the application domain by combining the three subjective interestingness measures: actionability, unexpectedness, and novelty. The model's feature selection phase optimally reduced the Dataset's features by 60% without compromising the model's performance. The experimental results show that the model gave a higher prediction accuracy (6% improvement for Dataset 1 and 7.4% improvement for Dataset 2) for the classification algorithms that produced the patterns. The SEM-AKD model demonstrates significant potential for real-world application where domain expert and multi-criteria pattern evaluation are critical.