Using phenotypic sharing in a classifier tool
Mozart Hasse, Aurora Pozo · 2000
This paper describes a classifier tool that uses a genetic algorithm to make rule induction. The genetic algorithm uses the Michigan approach, is domain independent and is able to process continuous and discrete attributes. Some optimizations include the use of phenotypic sharing (with linear complexity) to direct the search. The results of accuracy are compared with other 33 algorithms in 32 datasets. The difference of accuracy is not statistically significant at the 10% level when compared with the best of the other 33 algorithms. The implementation allows the configuration of many parameters, and intends to be improved with the inclusion of new operators.