Forecasting different exchange rates and commodity prices with machine learning approaches

The steel industry is widely acknowledged as a key indicator of economic progress owing to its significant contribution to social development and ubiquitous application in our lives. This underscores its pivotal role in determining our welfare. This thesis proposes three models using machine learning methods for price prediction in the steel sector. The first model utilizes all available data, the second model considers the last three years of data along with all foreign exchange rates and commodity prices to predict the next day's values, while the third model focuses on a specific model for Turkey's iron and steel prices.Upon examining the data obtained from three different prediction methods for 25 variables, it is observed that the Random Forest algorithm (RF) yields the best prediction results. RF achieves an absolute mean percentage error of less than 1% during training. Poor results are obtained with Artificial Neural Networks (ANN) at 19.6% and Support Vector Regression (SVR) at 300%. Furthermore, in the prediction study conducted by separating the last three years of the dataset, Random Forest Algorithm again demonstrates excellent performance, albeit with an increase in error rates for all models. Enhanced data set predictions result in more accurate outcomes. The analysis reveals a strong correlation between Turkey's CRC and HRC Turkey values and the dataset. For Turkey CRC, ANN produces a 1.67% error, RF a 0.55% error, and SVR a 2.63% error. For HRC Turkey, ANN yields a 1.71% error, RF a 0.57% error, and SVR a 2.35% error, indicating the model's accuracy and precision in generating reliable predictions. Finally, it is suggested that fine-tuning machine learning algorithms can lead to better results, and the optimization of hyperparameters has been observed to enhance prediction success. Particularly, optimizations for Turkey CRC and HRC Turkey dependent variables have resulted in more successful prediction outcomes for Artificial Neural Networks (1.49% - 1.52%), Random Forest Algorithm (0.49% – 0.52%), and Support Vector Machines (2.34% - 2.00%).