Generalized ATOVIC System Based on Triangular Fuzzy Numbers for Pattern Classification

Abstract

Diseases such as diabete and cancer must be detected and treated precociouse. They are generally detected using recognition and classification systems. The latter must be choosen according their accuracies and other measures. In addition, some difficulties and ambiguities can exist related with features representing data sets such as their variant types (eg., integer, nominal, etc.) or overlapping between them. Also, data can be imprecise or incomplete and can influence negatively classification results. Fuzzy sets and their generalizations, resolve problems of data uncertainty by modeling them with linguistic variables. The latter are presented with membership functions having different shapes. Triangular fuzzy numbers which represent tringular functions are very rarely used in classification systems. The cause can be that triangular shape does not preserve in case of some calculations. This research proposes to generalize the classification system "‘Amended Fused TOPSIS-VIKOR for classification"’ (ATOVIC) with triangular fuzzy numbers and to apply it to some UCI date sets. The objective is to unveil the impact of triangular fuzzy numbers in classification. The results of generalized fuzzy ATOVIC are compared to those of its crisp version and those of some existing classification systems using fuzzy sets. Results of ATOVIC generalized with tringular fuzzy numbers are promising. Especially its results on breast cancer Wisconsin and IRIS data sets are higher than Mamdani and Segano fuzzy inference systems using triangular fuzzy numbers from literature.