FORECASTING CURRENT EXCHANGE AND GOLD RATES WITH HYBRID MODELS USING TIME SERIES AND DEEP LEARNING ALGORITHMS

The flexibility and volatility experienced in exchange rates affect many financial activities in the world. In order to follow this situation, countries and multinational companies need to follow financial indicators in the world economy. There is a need for important decision systems that will follow all these systems with a reliable prediction model together with the developing technology. In this study, deep learning-based models that will accurately predict the movement of gold, dollar, and euro exchange rates are proposed. Time Series methods were used to analyze the data and make predictions. In addition to deep learning models such as LSTM, GRU, Bi-LSTM and RNN, hybrid models of these methods were also used to compare their prediction performances. The data set includes USD/TRY and EUR/TRY exchange rates and monthly prices of bullion gold in Turkish Lira. Data between the years 2000-2024 were included in the analyses, and a six-month future prediction of each rate was also made. In the study, the best results were obtained with a 98.39% f1 score in gold rate prediction with the GRU-RNN hybrid model. It was observed that hybrid models in particular provided higher accuracy in predictions in general. The findings show that optimizing model parameters has a significant impact on the success of financial forecasts.