The thesis explores the machine learning approaches for ischemia prediction based on ECG electrode data. We were interested in classiﬁcation accuracy at prediction of ischemia, prediction of pathological zones in heart, and possibility of reducing the number of atributes neccesary for successful detection. We used simulated data to train and test random forests, support vector machines and gradient boosting. We used these approaches to determine optimal attribute subsets using a wrapper approach, and compared how well methods perform on subsets of various sizes. We also compared the performance of our wrapper approach with a ﬁlter-based feature selection approach. Results show high classiﬁcation accuracy of all methods, even on small attribute subsets. Wrapper assisted support vector machines outperform other methods, and wrapper achieves better results than ﬁltering on small-sized subsets.