Mining A Primary Biliary Cirrhosis Dataset Using Rough Sets and a Probabilistic Neural Network

In this paper, a decision support system based on rough sets and a probabilistic neural network is presented. Rough sets were employed as they have the capacity to reduce the dimensionality of the dataset and also produce a set of readily understandable rules. A probabilistic neural network was also employed to classify this dataset, comparing the classification accuracy to that obtained with rough sets. We firstly evaluate the effectiveness of these machine learning algorithms on a real-life small biomedical dataset. The classification results indicate that both classifiers produce a high level of accuracy (87% or better). The rough sets algorithm produced a set of rules that are readily interpretable by a domain expert. The PNN algorithm produced a classifier that was robust to noise and missing values. These preliminary results indicate that the both rough sets and PNN machine learning approaches can be successfully applied synergistically to biomedical datasets that contain a variety of attribute types, missing values and multiple decision classes