2026-04-17T17:42:19.786Z http://ezaposleni.singidunum.ac.rs/rest/sciNaucniRezultati/oai

ezaposleni.singidunum.ac.rs/rest/sciNaucniRezultati/oai:3:9277 2024-11-10T17:06:24Z 3

Feature Selection and Extreme Learning Machine Tuning by Hybrid Sand Cat Optimization Algorithm for Diabetes Classification, Chapter in CCIS Communications in Computer and Information Science: MDIS 2022: Modelling and Development of Intelligent Systems, Springer, volume 1761 2023 http://ezaposleni.singidunum.ac.rs/rest/sciNaucniRezultati/oai/record/3/9277 https://link.springer.com/chapter/10.1007/978-3-031-27034-5_13 M. Stankovic N. Bacanin M. Zivkovic D. Jovanovic M. Antonijevic M. Bukumira I. Strumberger It is vitally important to establish a system that is able to provide an early detection of diabetes as a stealth disease of modern era. In order to achieve this goal, this manuscript proposes a novel framework for feature selection and extreme learning machine (ELM) hyper-parameter optimization applied to diabetes diagnostics. Feature selection and hyper-parameter optimization are two of the most important challenges in the domain of machine learning and they both belong to the group of NP-hard challenges. An upgraded version of the newly suggested sand cat swarm optimization (SCSO) is developed to address these issues, and adapted for ELM hyper-parameter tuning and feature selection. A preliminary set of biases and weights are established using the proposed approach, as well as the optimal (sub-optimal) no. of neurons for the ELM hidden layer, as well as to establish initial set of biases and weights. Furthermore, each swarm individual also tries to select the most relevant features for classification tasks against a widely utilized diabetes dataset. The performance of proposed methods was compared to other well-known state-of-the-art swarm intelligence algorithms in terms of accuracy, precision, recall and f1 score. Experimental findings demonstrate that the improved SCSO is more efficient than other algorithms in addressing both challenges. bookPart Springer, Cham 188 203 10.1007/978-3-031-27034-5_13 CCIS Communications in Computer and Information Science: MDIS 2022: Modelling and Development of Intelligent Systems, volume 1761