2026-05-09T11:15:13.779Z http://ezaposleni.singidunum.ac.rs/rest/sciNaucniRezultati/oai

ezaposleni.singidunum.ac.rs/rest/sciNaucniRezultati/oai:1:11208 2025-02-24T23:41:04Z 1

Modified Metaheuristic Optimization Algorithm for Hyperparameter Tuning in Mental Health Applications 2025 http://ezaposleni.singidunum.ac.rs/rest/sciNaucniRezultati/oai/record/1/11208 https://ieeexplore.ieee.org/abstract/document/10883383 S. Kozakijevic L. Jovanovic S. Tedic N. Jankovic M. Zivkovic N. Bacanin The World health organization considers a mentally healthy person to be able to manage life's stresses, work productively, and contribute to their communities. However, mental disorders, affecting 970 million people in 2019, often reduce quality of life, necessitating accurate diagnosis and therapy. Affective disorders, encompassing depressive and manic episodes, vary in severity and pattern, with conditions like bipolar disorder and recurrent mood disorders identified through ICD-10 classifications. Diagnosis, typically made by expert teams, remains challenging due to the complex and variable nature of mental disorders, leading to frequent misdiagnoses that hinder treatment progress and erode patient trust. Artificial intelligence (AI) holds promise in supporting mental health diagnosis by identifying subtle patterns in symptoms and validating expert conclusions. While AI's specialized nature poses limitations, integrating it into clinical practice can improve diagnostic accuracy and reduce errors. High quality models can assist doctors by ruling out less likely conditions and confirming diagnoses. Optimizing AI models, however, requires effective hyperparameter selection, a complex task classified as NP-hard. Metaheuristic algorithms offer a solution, enabling efficient parameter tuning and improving AI model performance. Given the critical implications of accurate mental health diagnoses, leveraging AI and optimization techniques is essential to enhance outcomes and improve lives. This work proposed a modified metahetusirc aimed at optimizing classification accuracy. When evaluated on a genuine dataset the best performing models attain an accuracy of 0.972222 suggesting feasibility in real world situations. conferenceObject 948 953 10.1109/ICMCSI64620.2025.10883383 2025 6th International Conference on Mobile Computing and Sustainable Informatics (ICMCSI), IEEE, Goathgaun, Nepal