Neural network based rule extraction for analysing risk factors and stages in cervical cancer—an analytical study

Abstract

The high performance and modelling ability of a Neural Network has enabled it to be used extensively in most domains, specifically in medical domain. Inspite of its excellent modelling performance, NN acts as a black-box because of its inability to provide a simple interpretation of the model. Analysing the network model is a challenging task. Several studies have been carried in this direction. In this paper we would like to propose a method for extracting knowledge from the model obtained by a multi layer perceptron. The rules extracted are precise and accurate. This method has been applied on cervical cancer data for its performance and found to produce accurate results. This technique is simple, efficient and comparable to the traditional rule based algorithms like J48. The results would be most valuable to the medical practitioners in diagnosing the disease at a very early stage.