LAUSR

This paper presents a new fuzzy modeling approach for analyzing censored survival data and finding risk groups of patients diagnosed with bladder cancer. The proposed framework involves a new procedure for integrating the frameworks of interval type-2 fuzzy logic and Cox modeling intrinsically. The output of this synergistic framework is a score/prognostics index which is indicative of the patient's level of mortality risk. A threshold value is selected whereby patients with risk scores that are greater than this threshold are classed as high-risk patients and vice versa. Unlike in the case of black-box type modeling approaches, the paper shows that interpretability and transparency are maintained using the proposed fuzzy modeling framework. Two datasets are used to test the modeling accuracy of the elicited models. The first is an artificial dataset which has similar characteristics as in a typical survival data. The second relates to real-life bladder cancer data from which one requires a model that identifies the low-risk and high-risk patients and then recommends risk management decisions based on, predicted risk level, patient history and characteristics, disease pathology, and event times. The performance of the proposed framework is compared with the traditional Cox model, logistic regression as well as a nonlinear survival data modeling technique based on neural networks. This is the first time an attempt has been made to exploit the transparency advantages of fuzzy models and the principled statistical framework of the Cox model in order to identify risk groups and recommend risk management decisions from complex survival datasets. In both the artificial data and real data, the proposed modeling framework, although minimalistic, shows better generalization performances than the previously reported models against which the results were compared.