Assessing the Effectiveness of Machine Learning Techniques for Silver Price Prediction: A Comparative Study

Abstract

Silver is considered an important asset in terms of economic indicators and a valuable investment asset in terms of the markets. Therefore, determining silver prices is critically important for both national economies and investors. However, the non-stationary and non-linear nature of silver prices makes predicting price movements challenging. The methods used for predicting silver prices must be suitable for capturing these volatile and complex behavioral characteristics. The silver market can be influenced by other commodities and investment assets. Factors affecting silver prices, such as gold prices, Brent crude oil prices, the US Dollar index, the VIX index, and the S&P 500 index, can play a significant role. In this context, these variables have been used as inputs for predicting silver prices in the study. Three different models have been developed to predict the prices one, two, and three days ahead. These models have been predicted using four different machine learning methods: linear regression, support vector regression (SMOReg), k-nearest neighbors (k-NN), and random forest (RF). The results show that the random forest and k-NN methods exhibit the highest performance. The random forest achieves the highest accuracy in the first two models, while k-NN excels in the third model. Linear regression and SMOReg methods are less successful compared to the others. Consequently, it can be concluded that random forest and k-NN methods can be preferred for long-term predictions, and that these results may provide valuable insights, especially for investors and decision-makers.