Forecasting Sudan Gold Prices with a Hybrid Deep Learning Approach

As a crucial element of the financial market, the price creation process of gold futures has attracted the attention of academicians and financiers alike. However, given the time series data’s disorganized, chaotic, and non-stationary nature, predicting the price of gold futures is extremely difficult. This paper introduces a novel approach for forecasting gold prices by combining a Time-Series Mixer (TSMixer), Long Short Term Memory Neural Networks (LSTM), and Convolutional Neural Networks (CNN). Our proposed method addresses the challenges posed by characteristics of gold price time series data. Our model consists of three key components: a Mixer block for extracting temporal patterns, an LSTM component to leverage the sequential order of daily gold prices, and a CNN component to capture local patterns and abstract spatial features. By integrating these elements, our model offers a thorough method to tackle the difficulties involved in gold price forecasting, taking into account the data’s temporal and spatial patterns. Our study focused on Sudan Gold futures as our task and selected gold futures data from the Sudan Gold Refinery company from January 6, 2015, to December 22, 2022. Through comparative analysis with baseline methods, our experiments show that our model outperforms all other baseline models. These findings highlight the superior performance of our model in effectively predicting Sudan Gold futures.