Cascade Generalization and Complementary Neural Networks for Multiclass Classification

This paper presents a technique for solving multiclass classification problems. Two existing techniques are combined which are cascade generalization and complementary neural networks. The unification of these two techniques can increase the efficiency of classification. Three small datasets from UCI machine learning repository are tested in the experiment. These datasets are wireless indoor localization, user knowledge modeling, and alcohol QCM sensor. The proposed approach gives the average accuracy of 98.5%, 95.0%, and 96.4%, respectively, which are better than using individual techniques such as feedforward backpropagation neural network, complementary neural networks, and cascade generalization.