2026-05-09T11:14:54.646Z http://ezaposleni.singidunum.ac.rs/rest/sciNaucniRezultati/oai

ezaposleni.singidunum.ac.rs/rest/sciNaucniRezultati/oai:3:11211 2025-02-27T21:09:26Z 3

Exploring the Potential of Modified Metaheuristic Optimized Long Short-term Memory Neural Networks for Earthquake Magnitude Forecasting 2025 http://ezaposleni.singidunum.ac.rs/rest/sciNaucniRezultati/oai/record/3/11211 https://www.taylorfrancis.com/chapters/edit/10.1201/9781003601555-6 N. Bacanin L. Jovanovic M. Dobrojevic M. Zivkovic M. Antonijevic M. Salb Natural disasters have a significant impact on communities and countries, often causing countless amounts of property damage, destroying livelihoods, and often costing human lives. One major contributing factor to the devastating nature of natural disasters is their unpredictability. Disasters such as earthquakes are additionally known for causing an aftershock following the main event, impacting rescue workers and survivors. A robust prediction system could help prevent the loss of human lives by providing a warning system for evacuations and preparations, as well as predicting aftershock occurrences, helping rescue and recovery workers plan accordingly. However, while a lot of data is available from seismology institutes around the world, this is not a simple task. This work examines the prospect of Long Short-term Memory (LSTM) for tackling earthquake magnitude prediction. Since the performance of LSTM models is highly dependent on adequate parameter selection, metaheuristic optimizers are applied to optimize performance and a new modified variant of the sine cosine algorithm is suggested to meet the demands of the introduced methodology. Several contemporary optimizers are included in a comparative analysis, with the proposed algorithm outperforming other optimizers. bookPart CRC Press 99 118 10.1201/9781003601555-6 Chapter in Multi-objective Optimization Techniques, 1st Edition