Application of genetic programming for multicategory pattern classification

Abstract

This paper explores the feasibility of applying genetic programming (GP) to multicategory pattern classification problem for the first time. GP can discover relationships among observed data and express them mathematically. Multicategory pattern classification has been done traditionally by using the maximum likelihood classifier (MLC). GP-based techniques have an advantage over statistical methods because they are distribution free, i.e., no prior knowledge is needed about the statistical distribution of the data. GP also has the ability to automatically discover the discriminant features for a class. GP has been applied for two-category (class) pattern classification. In this paper, a methodology for GP-based -class pattern classification is developed. The given -class problem is modeled as two-class problems, and a genetic programming classifier expression (GPCE) is evolved as a discriminant function for each class. The GPCE is trained to recognize samples belonging to its own class and reject samples belonging to other classes. A strength of association (SA) measure is computed for each GPCE to indicate the degree to which it can recognize samples belonging to its own class. The higher the value of SA, the better is the ability of a GPCE to recognize samples belonging to its own class and reject samples belonging to other classes. The SA measure is used for uniquely assigning a class to an input feature vector. Heuristic rules are used to prevent a GPCE with a higher SA from swamping a GPCE with a lower SA. Experimental results are presented to demonstrate the applicability of GP for multicategory pattern classification, and the results obtained are found to be satisfactory, and are compared with those of the MLC. We also discuss the various issues that arise in our approach to GP-based classification, such as the creation of training sets, the role of incremental learning, and the choice of function set in the evolution of GPCEs, as well as conflict resolution for uniquely assigning a class.

A coevolutionary algorithm for rules discovery in data mining

“…Ishibuchi et al (2001) tested this algorithm on several datasets, but only the training accuracy was provided for the Iris data. The GPCE was proposed by Kishore et al (2000), which is a GP-based technique dedicated to solving multi-category pattern recognition problems. In this algorithm, the n-class problem was modelled as n two-class problems and GPCE was trained to recognize samples belonging to its own class and reject samples belonging to other classes.…”

A constrained-syntax genetic programming system for discovering classification rules: application to medical data sets

“…The function set consists of logical operators (AND, OR) and relational operators ("=", "≠", "≤", ">"). [2,16]. In the i-th (i=1,...,k) run, the GP discovers rules predicting the i-th class.…”

Genetic Programming-Evolved Spatio-Temporal Descriptor for Human Action Recognition

“…Davis et al [3] have also employed GP for feature selection in multivariate data analysis, where GP can automatically select a subset of the most discriminative features without any prior information. In addition, other researchers [6,13,36] have also successfully applied GP to classification tasks with improvements compared with previous methods.…”