AbstractView references

It has always been a well-known fact that gold never loses its value. This makes gold a favorite for long-term investments both for individuals as well as countries. Being able to anticipate changes in gold price can be a very lucrative. However, as prices are influenced by many complex factors, casting accurate forecasts is not an easy task. The development of a robust and reliable method for casting accurate forecasts is evident. This work proposes an approach revolving around bidirectional long short-term memory neural networks optimized via metaheuristic algorithms to ameliorate performance. Additionally, a modified version of the well-known moth flame optimizer (MFO) algorithm is introduced for this purpose. To help the network deal with subtle complexities as well as violent variations and noise associated with market data, a decomposition techniques is applied to the univariate time series prior to network processing. The proposed approach and introduced algorithm have been evaluated on realistic data, and their performance has been collated to several trailblazing algorithms applied to the same task. The approach has shown promising results attaining the highest overall objective function scores, with overall MSE value of 0.000643. © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2024.