Computational method to prove efficacy of datasets

Abstract

Intrusion detection is one of the most significant area of research in sensor networks. Numerous machine-learning models have made a revolution in the domain of intrusion detection. Each machine-learning model differs in accuracy when authenticated with different dataset. An appropriate dataset may give a better accuracy as compared to an inappropriate dataset. In this paper, we have used three different dataset: KDDCup99 dataset, NSL-KDD Dataset and WSN-DS dataset for finding the accuracy of five most preferred machine learning algorithms: Decision Tree, K-Nearest Neighbor, Naïve Bayes, Random Forest, Support Vector Machine. The purpose of the research is to find out whether a new dataset WSN-DS gives a better accuracy as compared to existing datasets on same machine learning algorithms. Results prove that WSN-DS dataset outperforms with an accuracy of 99.64% than the NSL-KDD dataset and KDD-Cup99 dataset with an accuracy of 99.46% and 99.07% respectively, thus making it one of the best dataset available in the market.

Subject Classification: (2010):

• 68T05

Keywords:

• Intrusion detection
• Algorithm
• KDDCUP99 Dataset
• NSLKDD Dataset
• WSN_DS Dataset