Porovnání modelů strojového učení

The diploma thesis is focused on the comparison of machine learning methods for the regression task. The methods considered include: linear regression, shrinkage methods (lasso and ridge regression), tree-based methods (regression tree CART, random forest and bagging). The aim of this work is to examine the effectiveness, uniqueness of each method and compare the methods based on the ease of interpretation of models, quality of prediction, computational complexity and time. The diploma thesis discusses the properties and limitations of each method, the stability of models in the presence of multicollinearity, outliers and redundant information. The results of the model are supported by visualizations that these methods allow. All models are based on real data taken from the open source UC Irvine Machine Learning Repository. The work is focused on compiling a complete guide describing the construction of models, analysis of the advantages and disadvantages of the considered methods and analysis of the methods behavior in situations that affect the quality of models.