A modified projection based learning algorithm for a meta-cognitive radial basis function classifier

In this paper, we propose a modified Meta-Cognitive Radial Basis Function Network (McRBFN+) and its Projection Based Learning (PBL) algorithm for classification problems. During learning, as each sample is presented to McRBFN+, the modified meta-cognitive component monitors the prediction error and class-wise significance in cognitive component (RBFN) to efficiently decide on what-to-learn, when-to-learn and how-to-learn. The what-to-learn action is realized by sample-deletion-strategy, wherein samples with similar information content as the network are deleted without being learnt. The how-to-learn action realized by sample-learning-strategy decides on addition of new rule or update of existing neurons. A few samples not satisfying either what-to-learn or how-to-learn are reserved to be considered for learning at a later time by when-to-learn action. The sample-learning-strategy employs a PBL learning algorithm to evolve the structure and adapt the parameters of McRBFN+. The aim of PBL algorithm is to find the optimal output weight such that the sum of squared prediction error is minimized. During rule addition, meta-cognitive component performs judgement-of-learning to monitor and avoid any retrospective knowledge corruption. This helps the network avoid over-training as well as ensure knowledge is learnt efficiently. The performance of PBL-McRBFN+ classifier is evaluated on a set of benchmark classification problems from UCI machine learning repository. The performance comparison with existing methods indicate promising results.