Μοντέλα ταξινόμησης και εφαρμογές

Δοντάς, Γιώργος Ι.

Master Thesis

Author

Δοντάς, Γιώργος Ι.

Date

2012-10-08

Abstract

The aim of machine learning is to develop algorithms capable of improving their own performance, exploiting existing data, stored in huge databases, in order to discover knowledge and interpret several phenomena. Supervised learning aims in creating a model that takes into account the knowledge adapted by experience, and then uses it for evaluating new observations. One of the most common methods for describing phenomena is through classification, where a particular object is classified to one of several available classes of objects. The present thesis focuses on one of the most promising classification algorithms in the field of machine learning, the "support vector machine" (SVM). The presentation of the theoretical foundation advances gradually, starting from the most intuitive classification algorithm and reaching up to the optimized approach of SVM, so that it's easier for the reader to follow. During the presentation procedure, another two of the most popular classification algorithms are also highlighted: the "regularized logistic regression" and the "multi-layer perceptron". Beyond theoretical approach, this thesis aims developing appropriate algorithms, when possible, or otherwise to suggest how to use "of the shelf' and open-source software libraries. All the data used for the examples, as well as the whole of the implemented code, are available to the reader for experimentation. A machine learning process cannot be considered complete without having evaluated the model developed. For this reason, in the last chapter, we deemed it necessary to present several diagnostic tests and practical advice for model evaluation and optimization.