Selecting attributes for soft-computing analysis in hybrid intelligent systems

Abstract

Use of medical survival data challenges researchers because of the size of data sets and vagaries of their structures. Such data demands powerful analytical models for survival analysis, where the prediction of the probability of an event is of interest. We propose a hybrid rough sets intelligent system architecture for survival analysis (HYRIS). Given the survival data set, our system is able to identify the covariate levels of particular attributes according to the Kaplan-Meier method. We use 'concerned' and 'probe' attributes to investigate the risk factor in the survival analysis domain. Rough sets theory generates the probe reducts used to select informative attributes to analyze survival models. Prediction survival models are constructed with respect to reducts/probe reducts. To demonstrate the utility of our methods, we investigate a particular problem using various data sets: geriatric data, melanoma data, pneumonia data and primary biliary cirrhosis data. Our experimental results analyze data of risk factors and induce symbolic rules which yield insight into hidden relationships, efficiently and effectively. © Springer-Verlag Berlin Heidelberg 2005.