Using cellular evolution for diversification of the balance between accurate and interpretable fuzzy knowledge bases for classification

Recent work combining population based heuristics and flexible models such as fuzzy rules, neural networks, and others, has led to novel and powerful approaches in many problem areas. This study tests an implementation of cellular evolution for fuzzy rule learning problems and compares the results with other related approaches. The paper also examines characteristics of the cellular evolutionary approach in generating more diverse solutions in a multiobjective specification of the learning task, and finds that solutions seem to have useful properties that could enable anticipating out of sample performance. We consider a bi-objective problem of learning fuzzy classifiers that balance accuracy and interpretability requirements.