A Study on Learning of Ensemble Classification Model Based on Radial Basis Function Neural Networks

In this study, we introduce an ensemble model that leverages a radial basis function neural network as the base model to enhance the classification performance of a pattern classifier. While existing ensemble models use the majority vote technique for aggregation, we proposes a new aggregation method that combines a simple linear function and a softmax function, using the outputs of the base model as input variables. This aggregation approach linearly combines the class prediction probabilities from each base model to calculate the class-specific probability for the ensemble model. To evaluate the performance of the proposed pattern classifier, we used 14 datasets to optimize the parameters of the linear function and compared the performance with existing pattern classifiers. Our method achieved better results on 10 out of the 14 datasets compared to other pattern classifiers. Additionally, when compared with a recently reported ensemble model, our method also demonstrated superior performance.