An Extensive Analysis of Match-Tracking Methods for ARTMAP

This paper identifies and studies five match-tracking (MT) methods in the adaptive resonance theory (ART) literature and conducts a detailed comparative analysis of these in ARTMAP applications. We focus on model performance for each MT method with respect to time and space efficiency as well as classification accuracy. Experimental results indicate that one MT variant, used in ARTMAP applications for the first time in this work, provides significant improvements in computational efficiency: depending on the ARTMAP variant, it was able to achieve up to one order of magnitude reduction in both time and space requirements, albeit with a compromise in accuracy, relative to the most accurate MT variant. Furthermore, these results reveal the absence of a universally accuracy-optimal MT algorithm, suggesting that the optimal choice is dependent on the specific ARTMAP architecture and the dataset characteristics.