A hierarchical set-partitioning nonlinear discriminant classifier trained by an evolutionary algorithm
Ziauddin Ursani, David W. Come · 2018
This paper is the fourth in our series of papers on a novel nonlinear discriminant classifier. The model is based on elementary mathematical operators, yet is able to produce 100% accuracy on training data without compromising generalization ability. The reason behind the model's capabilities stem from its ability to partition the training data into a hierarchy of groups in such a way that a separate model is trained for each subgroup. Therefore, a Hierarchical Set-partitioning Nonlinear Discriminant Classifier is essentially a hierarchical arrangement of various models corresponding to hierarchical arrangement of various partitions/subgroups. In the latest version described here, the model has a low number of parameters, and we have extended its set-partitioning approach to include the concept of remainder unclassified group. These modifications have led to 100% results on the training set and highly competitive results on test sets when compared with the state of art on four popular test cases taken from the UCI repository.