Rank of normalizers through TOPSIS with the help of supervised classifiers

Abstract

Classification is a tedious task for gathering and categorizing collected knowledge from the noisy high-dimensional dataset. The classifier suffers a lot when the dimension of the dataset is high and the underlying dataset is in different size and units. To make the classification cost effective, the dataset must be subject to pre-processing. Normalization, as a pre-processor transforms the data into a unit less mode across all the dimensions of the dataset. Practically there are many normalization techniques which are best suitable for different implementation scenarios. Though, it is believed that normalization improves classifiers performance but it is a tedious task to ascertain an optimum normalizer for specific scenarios. In this paper, seven widely used normalization techniques are evaluated through five popular supervised learning classifiers using intrusion detection dataset. Rank to these normalization techniques are allocated using a popular ranking algorithm called as Techniques for Order Preference by Similarity to the Ideal Solution (TOPSIS), thus revealing the best optimum normalizer for intrusion detection environment.