Similar to neural network usage, a dataset with inputs and output pre-classified, and this generates not a model but a perl code that satisfies the outputs for the given inputs.
https://github.com/sha0coder/zerglin

** Generation 70  population size 100 error 9
9 E0,B3,D4-E4,B1,E3,A0,E5-E5,B5,E4-E2,B0,E8,A6,E1-E1,B0,E9,A5,E1-E4,B6
+,E3
$a*=4;$e+=$d+$f;$f**=$e;$c=$i**$b;$b=$j^$b;$e%=$d;

** Generation 71  population size 100 error 9
9 E0,B3,D4-E4,B1,E3,A0,E5-E5,B6,E4-E2,B0,E1,A6,E0-E4,B0,E9,A5,E1-E9,B6
+,E8
$a*=4;$e+=$d+$f;$f%=$e;$c=$b**$a;$e=$j^$b;$j%=$i;

** Generation 72  population size 100 error 4
4 E0,B1,E0,A2,E1
$a+=$a*$b;

** Generation 73  population size 100 error 4
4 E0,B1,E0,A2,E1
$a+=$a*$b;

** Generation 74  population size 100 error 4
4 E0,B1,E0,A2,E1
$a+=$a*$b;

** Generation 75  population size 100 error 4
4 E0,B1,E0,A2,E1
$a+=$a*$b;

** Generation 76  population size 100 error 0
0 E0,B0,E0,A2,E1
$a=$a*$b;

Optimized result:
E0,B0,E0,A2,E1
$a=$a*$b;
[download]

just for fun, there are many things to improve. regards.