NEURAL NETWORK CLASSIFICATION WITH PRIOR KNOWLEDGE FOR ANALYSIS OF BIOLOGICAL DATA

Neural Networks are efficient classification tools that have been applied to several applications including extracting regularities in data and classifying events in finance, marketing, internet and biomedicine. The training process uses available examples to produce a model and classify new events based on the extracted model. This learning procedure based on the examples is not capable of taking prior knowledge that is either available or discovered in data into account. In the present work, we propose a way to include prior knowledge into Radial Basis Function Neural Networks and to express the knowledge as a set of linear constraints in the least square problem. The obtained method still takes advantage of kernel functions to obtain a nonlinear classifier. Furthermore, its computational complexity is not affected while the misclassification error is enhanced. Publicly available biomedical datasets are used in a case study to analyze the performance of the approach, and to compare the results with the state of the art classifiers.