Includes references (73-83 p.)

'ABSTRACT Over the decades, ensuring security of networks has been a major concern; the open-ended nature of networks though promoting its ease of use and connectivity has also contributed its security challenges. The growth in technology fosters the developments of varies gadgets but it is also being harnessed by intruders to develop new and more sophisticated ways of breaking into a network, hence the need of more sophisticated methods that can detect new day attacks in a various types of networks. Different algorithms have been used in building network security systems, an example of one and what this work will be focusing on is a classification algorithm- minimum distance classifier. However in spite of effectiveness in ensuring network security, these algorithms also have different disadvantages that reduce its performance and a major way of improving the performance is through feature selection. This research focuses on how feature selection methods can be used to improve the performance of these algorithms; the research enumerates the importance of anomaly detection in network security, it also examines different past works that have used feature selection methods to improve different types of algorithms, it then execute the experiment using minimum distance classifier on a UNBIDS 2012 dataset. The performance of the classifier is optimized using two basic feature selection methods: entropy and variance, then a k-fold cross validation is performed to validate the accuracy results. Keywords: feature selection, lazy learning, non-parametric, minimum distance classifier, entropy, variance, and cross-validation.'