A Hybrid Connectionist-Symbolic Approach for Real-Valued Pattern Classification

Knowledge-Based Artificial Neural Networks (KBANNs) offer a means for combining symbolic and connectionist approaches into a hybrid methodology capable of dealing with small datasets and requires shorter training time when compared to conventional artificial neural networks (ANNs). These approaches were developed mainly for binary-valued domain problems and when applied to real-valued data, many tedious transformation processes need to be undertaken before it could be used. In this paper we propose to extend KBANN to a Real Knowledge-Based Neural Network (RealKBNN) that is able to handle real-valued data directly, making it a more powerful and realistic methodology to be applied to real-world problem domains. Four realvalued datasets (one artificial, 2 medical and 1 classification datasets) were used to evaluate RealKBNN. The results showed that RealKBNN performs very well in terms of its ability to correctly classify unseen examples as well as its suitability for classification of scarce and complex real data.