2026-05-04T18:31:05.435Z http://ezaposleni.singidunum.ac.rs/rest/sciNaucniRezultati/oai

ezaposleni.singidunum.ac.rs/rest/sciNaucniRezultati/oai:2:11801 2026-01-16T19:34:20Z 2

Unemployment Multivariate Forecasting: An Approach Combining Long Short-Term Memory Network and Modified Particle Swarm Optimization Algorithm 2026 http://ezaposleni.singidunum.ac.rs/rest/sciNaucniRezultati/oai/record/2/11801 https://link.springer.com/article/10.1007/s10614-026-11313-y J. Cadjenovic Milovanovic L. Babic D. Jovanovic M. Zivkovic L. Jovanovic M. Antonijevic M. Abdel-Salam N. Bacanin Monitoring unemployment trends is crucial as they provide a key indicator of an economy’s overall health, enabling policymakers, governments, and central banks to assess economic conditions and implement effective policies when needed. These trends also have profound social implications, impacting public health, societal well-being, and social stability. Elevated unemployment rates are often associated with increased poverty, social inequality, and mental health challenges. Furthermore, unemployment trends serve as predictors of future economic performance, highlighting the importance of accurate forecasting for policymakers, economists, and businesses worldwide. Traditional forecasting methods often fall short in identifying complex, non-linear patterns within data, limiting their predictive accuracy. Machine learning (ML) techniques offer a promising alternative due to their ability to process large datasets, uncover hidden relationships, and adapt to dynamic economic conditions. By leveraging advanced algorithms, ML-based approaches can significantly enhance the precision of unemployment rate predictions. Therefore, this study introduces a hybrid approach for unemployment forecasting, leveraging a long short-term memory (LSTM)-based model with attention mechanisms. To optimize the model’s performance, a variant of the particle swarm optimization (PSO) algorithm was employed to fine-tune its hyperparameters. Simulations conducted on real-world data, along with rigid comparative and statistical analysis with other methods, demonstrate promising results, underscoring the approach’s potential for far-reaching socio-economic benefits. article 10.1007/s10614-026-11313-y 1 31 1572-9974 Computational Economics