Impact of Missing Data Imputation on Intrusion Detection Performance: An Empirical Evaluation on NSL-KDD

Abstract

 For reliable detection performance, intrusion detection system uses machine learning algorithm for converting missing value data to superior quality.becauses of daily traffic on network dataset contain some values that are missing due to sensor failure, packet loss or concern about privacy. This makes intrusion detection system less precise. Often some research neglect data missing and some other use basic techniques to deal with it without exploring. In this paper we examine all thevarious missing data mechanism like MCAR, MAR and MNAR mechanism on NSL-KDD dataset. We also use variety of missing value imputation strategies such as stastical and machine learning based approaches. Based on result, Random Forest-based imputation significantly improves accuracy and the F1-scores under all missing situation. The findings highlight the importance mechanism-aware missing data treatment is for effective intrusion detection.