Analysis Of The Financial Performance Of Machine Learning Models For Predicting The Direction Of Changes In Cee And See Stock Market Indices With Different Classification Evaluation Metrics

The aim of the analysis is to investigate the influence of the selection of classification evaluation metrics on the financial performance of trading systems based on machine learning models for stock market indices from CEE and SEE regions. Technical indicators are used as features for selected machine learning algorithms when predicting the direction of index value changes, i.e. classifying trading days into two classes. The research showed that the choice of classifier evaluation metrics does not have a great impact on the financial performance of such a system. However, the highest average returns per trade were achieved by maximizing accuracy. Furthermore, the random forest algorithm and the naive Bayesian classifier gave the highest average returns using accuracy, while the support vector machine and the k-nearest neighbor algorithm achieved the highest average returns when using the area under the receiver operating characteristic curve. It was determined that the choice of machine learning algorithm has an expectedly large impact on financial performance and that the random forest algorithm gives the best results on this data.