Abstract:

Aiming at the unnecessary need of manual input of parametersτ1 andτ2, we propose a sementic driven subtractive clustering mehtod (SDSCM) to improve the traditional subtractive clustering algorithms by introducing the axiomatic fuzzy sets (AFS) thoery. The density radius τ1 is automatically determined and weight τ2 is semi-automatically determined based on the membership matrix. We compare the SDSCM with the FCM and the KMEANS on the Iris and Wine data sets. Experimental results show that the SDSCM is 1% to 5% higher than the FCM and the KMEANS in terms of evaluation index (semantic strength expectation), while the SPT is lower than the other two methods, where it still needs to be improved. The SDSCM can effectively solve the disadvantages caused by manual input of parameters τ1 and τ2, and find the clusters which are closer to the user semantics.

Key words: subtractive clustering, AFS, membership matrix, semantic strength expectation