Maskinlæringsteknikker for klassifisering

This diploma thesis treats classification tasks in machine learning. It focuses on three types of classification algorithms: support vector machines, neural nets and decision trees. The scope of its treatment of neural nets is limited to feedforward nets (i.e. neural nets without internal feedback loops), trained by backpropagation. Several types of decision trees exist. This thesis focuses on an algorithm called C4.5. As a part of the thesis, a classification module was written in VisualWorks 3.0 Smalltalk. The module consists of three submodules, each implementing one of the classification techniques mentioned. The three modules were tested on non-trivial examples. Some of the experiments showed that the classifiers were able to learn the training data perfectly, and that their parameters could be set to achieve good generalisation as well.