Type-2 Mamdani Fuzzy System Optimization for a Classification Ensemble with Black Widow Optimizer

Ensemble models perform classification on data using multiple models combined and unified to a single output. An important part of the ensemble is the combination or aggregation process. Even so the design of the aggregator for each ensemble could be very different depending on the problem. For this case, a proposed aggregation algorithm combines two classification submodels using a Mamdani fuzzy system. Depending on the data to learn, the fuzzy system could be designed differently. To solve this, we applied a Bio-inspired optimization algorithm, the Black Widow Optimizer, to adjust the fuzzy system into the data the ensemble learns. Adjustment to the problem is done by obtaining the best fuzzy system parameters including membership function points, type of function and optimizing the fuzzy rules. The optimization is compared with other optimization algorithms and experiments for the ensemble are done on two classification datasets of medical images.