Abstract
Medical diagnostic system is a branch in bioinformatics that is concerned with classifying medical records. Breast cancer is the most common deployed cancer in females worldwide. The main obstacle is the vagueness and ambiguity involving the breast cancer data. Human nature handles the vagueness and ambiguity easily. Therefore, doctors diagnose the patient condition using their expertise. Fuzziness and rough boundary theories simulate the human thinking. The fuzzy rough hybrids address the uncertainty in terms of membership degree of truth and lower and upper boundaries of fuzzy rough set theory. This research solves the diagnostic breast cancer problems via a proposed hybrid model of fuzzy rough feature selection and rough neural networks. The medical data is preprocessed by the fuzzy rough feature selection algorithm to remove unnecessary attributes. The reduced data set is applied to the rough neural network to learn the connection weights iteratively. The test data set are used to measure the proposed model accuracy and time complexities. Lower and upper approximations of the input features are weighted by input synapses learnt through training phase. The fuzzy rough proposed model design and implementation are declared. The experiments used WDBC and WPBC data sets from the UCI machine learning repository. The experimental results proved the fuzzy rough model ability to classify new instances compared with the conventional neural network.