Applying Clustering Based on Rules for Finding Patterns of Functional Dependency in Schizophrenia

In 1996 Fayyad described the Knowledge Discovery process as an integral process including prior expert knowledge, preprocessing, data mining and knowledge production to produce understandable patterns from data. Clustering based on rules (ClBR) is a particular data mining method suitable for profiles discovery. ClBR is an hybrid AI and Statistics technique, which combines some Inductive Learning (from AI) with hierarchical clustering (from Statistics) to extract knowledge from complex domains in form of typical profiles. It has the particularity to embed the prior expert knowledge existent about the target domain in the clustering process itself, guaranteeing more comprehensible profiles. In this paper, the results of applying this technique to a sample of patients with mental disorders are presented and their advantages with regards to other more classical analysis approaches are discussed. The final step of knowledge production is supported by post-processing tools, like Class panel graphs (CPG) and Traffic Lights panels (TLP), which were appreciated by domain experts as powerful, friendly and useful tools to transform raw clustering results into understandable patterns suitable for later decision-making. It was confirmed that functional impairment (FI) in schizophrenia and other severe mental disorders show a different pattern than FI in physical disability or in ageing population. Understanding the patterns of dependency in schizophrenia and getting criteria to recognize them is a key step to develop both elegibility criteria and services for functional dependency in this particular population. This research was related with the implantation of the Spanish Dependency Low, in Catalonia, acting from 2007.