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

Abstract

This paper proposes a new constrained-syntax genetic programming (GP) algorithm for discovering classification rules in medical data sets. The proposed GP contains several syntactic constraints to be enforced by the system using a disjunctive normal form representation, so that individuals represent valid rule sets that are easy to interpret. The GP is compared with C4.5, a very well-known decision-tree-building algorithm, and with another GP that uses boolean inputs (BGP), in five medical data sets: Chest pain, Ljubljana breast cancer, Dermatology, Wisconsin breast cancer, and Pediatric Adrenocortical Tumor. For this last data set a new preprocessing step was devised for survival prediction. Computational experiments show that, overall, the GP algorithm obtained good results with respect to predictive accuracy and rule comprehensibility, by comparison with C4.5 and BGP.

GEPCLASS: A Classification Rule Discovery Tool Using Gene Expression Programming

“…GEPCLASS used the following parameters: population size: 30 individuals; stopping criterion: 50 generations; number of genes per chromosome: 2; linking function: logical and; head size range: 6-10; selection method: stochastic tournament; genetic operators: all those defined by the original GEP [3] with the same probabilities. All results reported in this work were obtained by performing a 5-fold cross-validation procedure [5], and using exactly the same data partitions as in [2]. Table 1 shows the accuracy rate for the four data sets, using C4.5, CSGP, BGP, and GEPCLASS.…”

“…Therefore, we compared GEPCLASS with a constrainedsyntax genetic programming (CSGP) system proposed by [2], over four realworld data sets. In that work there is also a comparison with a "Booleanized" version of genetic programming (BGP) [1] that we reproduced here.…”

“…The CSGP used in [2] had the following parameters: maximum tree depth: 15; population size: 500 individuals; stopping criterion: 50 generations; recombination, reproduction and mutation operators probabilities: 95%, 5% and 0%. GEPCLASS used the following parameters: population size: 30 individuals; stopping criterion: 50 generations; number of genes per chromosome: 2; linking function: logical and; head size range: 6-10; selection method: stochastic tournament; genetic operators: all those defined by the original GEP [3] with the same probabilities.…”

“…An extension of this rule is the implicit precedence between functions (logical and relational) and terminals. In practice, this is accomplished by means of a constrained syntax, inspired in [2]. There is, the encoding must guarantee that a given operator will receive valid operands (terminals), and, therefore, A consequence of the filling rule is regards to the nature of the attributes.…”

Applications of genetic programming in cancer research

“…In (Bojarczuk 2004), a number of datasets were analyzed using genetic programming, including two breast cancer datasets and an adrenocortical cancer dataset. The first breast cancer dataset is the Wisconsin breast cancer dataset described in the UC Irvine Machine Learning Data Repository (MLDR).…”