Learning with imbalanced datasets using fuzzy ARTMAP-based neural network models

One of the main difficulties in real-world data classification and analysis tasks is that the data distribution can be imbalanced. In this paper, a variant of the supervised learning neural network from the Adaptive Resonance Theory (ART) family, i.e., Fuzzy ARTMAP (FAM) which is equipped with a conflict-resolving facility, is proposed to classify an imbalanced dataset that represents a real problem in the semiconductor industry. The FAM model is combined with the Dynamic Decay Adjustment (DDA) algorithm to form a hybrid FAMDDA network. The classification results of FAM and FAMDDA are presented, compared, and analyzed using several classification metrics. The outcomes positively indicate the effectiveness of the proposed FAMDDA network in undertaking classification problems with imbalanced datasets.