Mining clusters and corresponding interpretable descriptions - a three-stage approach

This paper presents a three-stage approach to data mining which puts special emphasis on the visualization and interpretability of the results. In the first stage, the input data is represented by a selforganizing map in order to allow visualization and to reduce the amount of data while removing noise, outliers, and missing values. Then this preprocessed information is used to identify and display fuzzy clusters of similarity. Finally, descriptions close to natural language are computed for these clusters in order to provide the analyst with qualitative information. This is accomplished by generating fuzzy rules using an inductive learning method. The proposed approach is applied to three case studies, including image data and real-world data sets. The results illustrate the robustness, intuitiveness and wide applicability of the method.