Agnieszka Nowak-Brzezińska

This paper presents a novel approach to optimize inference in rule-based knowledge systems by introducing a clustering mechanism for rule organization. Rules are clustered using K-Means or Agglomerative Hierarchical Clustering (AHC) algorithms, with different distance measures and clustering strategies. We propose and evaluate four inference strategies based on different group representation methods (mean or median) and rule activation strategies (activation of one or all matching rules). Experimental studies on real knowledge bases show that clustering significantly improves inference performance while maintaining a satisfied inference success rate.

Unsupervised models are becoming increasingly common in business processes. They are extremely effective in cases where we don't have a clearly defined decision class or the data contains anomalies that are hard to identify. The problem emerges in the effective processing of categorical data. Recently, many new approaches have been designed to analyze this data type. Still, most of them do not address the issue of unbalanced datasets, which is extremely difficult to catch when dealing with unlabeled data. Moreover, it is sometimes challenging to determine when abnormal observations represent a small data cluster and when they are already anomalies. This research analyzes several less popular algorithms that solve this problem and automatically place abnormal observations into separated clusters. We have shown that such methods are much better at clustering unbalanced data but also perfectly detect outliers in categorical datasets.

Knowledge engineering and data mining are the two biggest pillars of modern intelligent systems