Prototypes stability analysis in the design of fuzzy classifiers to assess the severity of scoliosis

The purpose of this paper was to develop and test a fuzzy classifier system to assess and monitor the severity of scoliosis. To design a reliable fuzzy classifier system, a notion of prototypes stability was introduced. Prototypes, which can be seen as representatives of information granules, need to be stable (i.e., they should not differ significantly in spite of small fluctuations occurring within the experimental data). If they are stable, prototypes could be used in the design of different learning architectures. In this work, prototypes stability analysis was used to find the number of clusters (or information granules) appropriate for classifier design. Once the number of clusters was found, a fuzzy relational classifier was designed and fuzzy rules were extracted. The usefulness of the proposed method was illustrated with the aid of numeric studies including two well-known datasets and a database of patients with scoliosis.