Cluster-dependent rotation-based feature selection for the RBF networks initialization

The paper addresses the problem of the radial basis function network initialization with feature section carried-out independently for each hidden unit. In each case a unique subset of features is derived from respective clusters of instances using the rotation-based ensembles technique. The process of the RBFN design with cluster-dependent features, including initialization and training, is carried-out using the agent-based population learning algorithm. The approach is validated experimentally and the obtained results are compared with the results produced using other methods.