Comparative analysis of machine learning models for cryptocurrency price prediction and selection of the optimal method using MCDM techniques

In recent years, cryptocurrencies have become a major focus of attention and are widely regarded as investment opportunities. However, due to the highly volatile nature of the cryptocurrency market, price prediction has become a challenging task. In this context, the aim of this thesis is to compare different machine learning models for predicting cryptocurrency prices and to identify the most effective method using Multi-Criteria Decision Making (MCDM) techniques. The study integrates historical price movements of cryptocurrencies, technical indicators, macroeconomic variables, and investor sentiment indicators. Prediction models were developed using SVM, LSTM, RF, and XGBoost algorithms, based on daily data from the period 2018–2023. The performance of the models was evaluated using statistical error metrics such as R², MAE, MSE, and RMSE. In addition, the importance of variables was analyzed using permutation importance for SVM and LSTM, and embedded feature importance methods for RF and XGBoost. In the final stage, in order to determine the best method via MCDM techniques, the CRITIC method was used to assign weights to the criteria, and the TOPSIS, ARAS, and CODAS methods were employed for ranking. The overall ranking was obtained through the Copeland method. According to the results, all four methods performed successfully in predicting cryptocurrency prices. Among them, the XGBoost algorithm demonstrated the highest overall performance based on the Copeland method, followed by LSTM in second place, RF in third, and SVM in fourth.