Evolutionary Mining of Rule Ensembles

Ensemble rule based classification methods have been popular for a while in the machine-learning literature (Hand, 1997). Given the advent of low-cost, high-computing power, we are curious to see how far can we go by repeating some basic learning process, obtaining a variety of possible inferences, and finally basing the global classification decision on some sort of ensemble summary. Some general benefits to this idea have been observed indeed, and we are gaining wider and deeper insights on exactly why this is the case in many fronts of interest. There are many ways to approach the ensemblebuilding task. Instead of locating ensemble members independently, as in Bagging (Breiman, 1996), or with little feedback from the joint behavior of the forming ensemble, as in Boosting (see, e.g., Schapire & Singer, 1998), members can be created at random and then made subject to an evolutionary process guided by some fitness measure. Evolutionary algorithms mimic the process of natural evolution and thus involve populations of individuals (rather than a single solution iteratively improved by hill climbing or otherwise). Hence, they are naturally linked to ensemble-learning methods. Based on the long-term processing of the data and the application of suitable evolutionary operators, fitness landscapes can be designed in intuitive ways to prime the ensemble’s desired properties. Most notably, beyond the intrinsic fitness measures typically used in pure optimization processes, fitness can also be endogenous, that is, it can prime the context of each individual as well.