ICF: An algorithm for large scale classification with conic functions

Abstract

Incremental Conic Functions (ICF) algorithm is developed for solving classification problems based on mathematical programming. This algorithm improves previous version of conic function-based classifier construction in terms of computational speed. Furthermore, the incremental step avoids the a-priori knowledge of number of sub-classes (which is a necessary parameter in the clustering step of this classification algorithm). Test results show that ICF is, on the average almost 3-times faster than previous versions without sacrificing accuracy. Python 2.7 implementation and software explanations are provided. Keywords: Polyhedral conic functions, Mathematical programming, Classification, Machine learning