B. Tusor, A. R. Várkonyi-Kóczy, S. Gubo: Hyperparameter Tuning for Sequential Fuzzy Indexed Search Trees Classifiers for Biosignal Processing. In IEEE International Symposium on Medical Measurements and Applications (MeMeA 2025) Proceedings. pp. 1–6, 2025. ISBN 979-8-3315-2347-3 link

Abstract: Classification is an integral part of machine learning, that is widely used in numerous applications in many areas of the medical industry. In previous work, the authors proposed the so-called Sequential Fuzzy Indexed Search Trees classifier that combines fuzzy inference, indexing tables and self-balancing binary search trees to achieve a fast and accurate classification, which can be used for quickly classifying medical data as well, be it singular data or large datasets of biosignals. However, a deeper analysis has not been done yet regarding its hyperparameters and how they affect the classification performance of the classifier. In this paper, a novel hyperparameter-tuning algorithm is proposed for the SFIST classifier that is based on the K-fold cross-validation method. The proposed method operates with four possible modes that approach the tuning process from two different aspects, namely to whether or not set the hyperparameters for each attribute individually or uniformly, and to either put the focus on the hyperparameters or the attributes during the process. All four modes have been tested and evaluated on benchmark datasets to establish which one results in the most reliable hyperparameter values. Furthermore, the best performing settings are used to compare the overall classification performance of the SFIST classifier to that of two state-of-the-art classifiers, as well as to untuned SFISTs to measure the improvement from the tuning step.