Machine learning is one of the most interesting topics of research that is applied in many domains such as for example medicine. It is going to play an important role for decision support in this area. The amount and complexity of recorded data in this area increases constantly, which makes it harder for humans to make right decisions that are important for human lives. The focus of this thesis is the application of the machine learning techniques in medical data. An important question which arises, when applying machine learning techniques, is the selection of the most suitable techniques for a specific application. Although many researchers compared different techniques for specific medical domains, often it is not clear if the results for these domains can be still improved by applying other machine learning techniques. In this thesis, several medical data sets were selected from UCI repository. The focus was particularly on data sets for which is not easy to achieve high classification accuracy. In this thesis we first reviewed the machine learning techniques which have been used for the selected data sets and analyzed the existing results. We then experimentally evaluated various new classifiers on these data sets. The parameters in each classifier were investigated and experiments with various configurations were performed. Furthermore, we evaluated the impact of the preprocessing techniques on selected datasets. The experiments showed that the use of preprocessing techniques and parameter tuning is very important to achieve good performance for the most machine learning algorithms. Although the best results were obtained by various machine learning algorithms, our experiments showed that ensemble learning algorithms such as AdaBoost and random forest gave usually good results.