Data Drive Fuzzy Cognitive Map for Classification Problems

In recent years Fuzzy Cognitive Maps had become an important tool for expert knowledge representation due to the flexibility and interpretability of modeled maps. Its construction frequently requires an expert's intervention but, there are situations when only the data is available or is required to extract the contained implicit knowledge for analysis or decision making proposes. Several studies have been developed to improve or find causal relation values between the map concepts but usually require a previous concept definition step carried out by experts. The frequent pattern mining techniques show a way for non-trivial relations extraction from datasets, and those relations may represent a causality degree. In this paper, a strategy to extract concepts from continuous and discrete features for supervised classification problems is proposed. Additionally, to estimate the causality degree between defined map concepts is proposed to use association rule mining techniques. Finally, the strategy is evaluated to show the interpretability and accuracy of generated Fuzzy Cognitive Maps.