The unemployment rate is an important factor for numerous reasons. The obvious one is the economy, but on the other hand, it affects a lot of fields and is important for the advancement of civilization. Since this problem can be presented as a time-series prediction task, it can be solved with modern artificial intelligence (AI) techniques such as recurrent neural networks (RNNs). Furthermore, additional performance improvements are possible through the application of an attention mechanism in such RNNs. However, the main problem with the use of such techniques is hyperparameter optimization, which is needed to produce the maximum performance of an RNN. This can be performed by another algorithm that is integrated into the proposed system. In this manuscript, a modified version of the chimp optimization algorithm (ChOA), a metaheuristic-based optimizer from the group of swarm intelligence algorithms, is used for RNN with attention mechanism hyperparameters tuning. Significant improvements have been noted in comparison to the original solution, but the algorithm also outperforms other state-of-the-art metaheuristics for unemployment rate prediction task in terms of standard regression metrics making it suitable for similar real-world challenges.