Investigation of The Effect of Data Normalization on Classification and Feature Selection in Intrusion Detection System


  • Mesut Polatgil Sivas Cumhuriyet √úniversitesi



Intrusion Detection System , machine learning , feature selection , data scaling , data normalization


The increasing use of the internet and the developments in the world of informatics have brought security problems together. Hardware and software known as Intrusion Detection System aim to detect attacks from the outside world and protect the system from them. These systems need to be fast and intelligent. Establishing intelligent systems for IDS requires data collection, processing and establishment of models in this area. It is very important to pre-process the collected data and select the necessary attributes. The fact that there are many feature selection methods and data preprocessing steps raises the question of which of these should be used and even which of these combinations of options would be better. Although there are studies on selecting the required features or on different normalization methods, there is no study that applies them together on IDS systems. This study was carried out for this purpose. With the study performed on the Kddcup99 dataset, 4 different normalization and 4 different feature selection methods were evaluated together. In this context, 20 different datasets were created and K nearest neighbor, Artificial Neural Networks and Random Forest algorithms were applied to each of them. When the results obtained are evaluated together with the normalization methods and feature selection methods, it has been shown that ideal features that produce successful results for IDS can be found