Classification of spam with the LSNB method

The issue of spam is more and more concerning with the seemingly infinite growth of the internet. The aim of this bachelor thesis is the implementation of an algorithm based on the method Loosely symmetric naïve Bayes, for use in the classification of spam from small and biased datasets. Authors of the algorithm have found a discrepancy between the data that is used to train the classifier, and data that is found in practical applications. The LSNB model is created to deal with this discrepancy in a satisfying manner. The implementation is achieved with the Python programming language, whose extension scikit-learn is the basis for the implementation.The theoretical part presents an introduction to the issue of spam, then the general machine learning methods are introduced. But also, some of the specific methods are described in detail, in particular the naïve Bayes classifier. Afterwards, the theoretical LSNB model which attempts to use cognitive bias for reproducing human-level concept learning is described. Some data pre-processing techniques are introduced together with the basics of Python language and some of its most important extensions. Finally, the metrics for classifier evaluation are presented.In the practical part the implementation is described in detail from the data pre-processing, through model fitting from training data, to classification of testing data. Evaluation of 6 different classifiers on 6 datasets of various sizes and various biases is shown near the end. For this, the metrics described in the theoretical part are used. The eLSNB classifier, which has been implemented as a part of this thesis, has performed the best in comparison with the other classifiers. It is deemed suitable for further testing.