Class Balanced Similarity-Based Instance Transfer Learning for Botnet Family Classification

Date

2018

Advisors

Journal Title

Journal ISSN

ISSN

DOI

Volume Title

Publisher

Type

Conference

Peer reviewed

Yes

Abstract

The use of Transfer Learning algorithms for enhancing the performance of machine learning algorithms has gained attention over the last decade. In this paper we introduce an extension and evaluation of our novel approach Similarity Based Instance Transfer Learning (SBIT). The extended version is denoted Class Balanced SBIT (or CB-SBIT for short) because it ensures the dataset resulting after instance transfer does not contain class imbalance. We compare the performance of CB-SBIT against the original SBIT algorithm. In addition, we compare its performance against that of the classical Synthetic Minority Over-sampling Technique (SMOTE) using network tra ffic data. We also compare the performance of CB-SBIT against the performance of the open source transfer learning algorithm TransferBoost using text data. Our results show that CB-SBIT outperforms the original SBIT and SMOTE using varying sizes of network tra ffic data but falls short when compared to TransferBoost using text data.

Description

Keywords

Similarity based transfer learning, Botnet detection, Synthetic minority oversampling technique (SMOTE), Transferboost

Citation

Alothman, B., Janicke, H., and Yerima, S. Y. (2018) Class Balanced Similarity-Based Instance Transfer Learning for Botnet Family Classification. In: Proceedings of the 21st International Conference on Discovery Science, DS 2018, 29-31st October 2018, Limassol, Cyprus.

Rights

Research Institute