Automatic Generation of ANFIS Rules in Modelling Breast Cancer Survival

Abstract

Data collected to be processed by means of rules can be done in multiple ways. In our previous papers, the Adaptive Neuro-Fuzzy Inference System (ANFIS) has been applied to breast cancer data for modelling survival in the presence of censorship. In initial work, the membership functions for the input data were defined by experts, along with an estimation of output. However, if knowledge about the data is vague or the expert cannot express the knowledge explicitly, the initial membership functions can be defined by partitioning the input space equally. Extracting fuzzy rules from the data using clustering methods is another technique used to initialise the position of membership functions of the input data. In this paper, we investigate whether such automatic methods can be used to initialise the antecedents of our model. Two clustering methods were applied to partition the input space, namely fuzzy c-means clustering and subtractive clustering, to establish the initial membership functions and a set of rules for the models. Further, to improve the model performance and high model accuracy, model optimisation was performed using the ANFIS approach.