Microsoft Malware Classification Challenge

The Microsoft Malware Classification Challenge was announced in 2015 along with a publication of a huge dataset of nearly 0.5 terabytes, consisting of disassembly and bytecode of more than 20K malware samples. Apart from serving in the Kaggle competition, the dataset has become a standard benchmark for research on modeling malware behaviour. To date, the dataset has been cited in more than 50 research papers. Here we provide a high-level comparison of the publications citing the dataset. The comparison simplifies finding potential research directions in this field and future performance evaluation of the dataset.

READ FULL TEXT VIEW PDF
POST COMMENT

Comments

There are no comments yet.

Authors

page 1

page 2

page 3

page 4

11/13/2015

Novel Feature Extraction, Selection and Fusion for Effective Malware Family Classification

Modern malware is designed with mutation characteristics, namely polymor...
11/22/2021

A Comparison of State-of-the-Art Techniques for Generating Adversarial Malware Binaries

We consider the problem of generating adversarial malware by a cyber-att...
01/21/2021

Malware Detection and Analysis: Challenges and Research Opportunities

Malwares are continuously growing in sophistication and numbers. Over th...
08/27/2019

A characterisation of system-wide propagation in the malware landscape

System-wide propagation is frequently observed in malware, and there are...
09/13/2019

On educating machines

Machine education is an emerging research field that focuses on the prob...
11/29/2021

MOTIF: A Large Malware Reference Dataset with Ground Truth Family Labels

Malware family classification is a significant issue with public safety ...
08/01/2019

KiloGrams: Very Large N-Grams for Malware Classification

N-grams have been a common tool for information retrieval and machine le...

Code Repositories

KaggleMicrosoftMalware

Repository for team's work on the Microsoft Malware Classification Challenge Kaggle competition.


view repo
This week in AI

Get the week's most popular data science and artificial intelligence research sent straight to your inbox every Saturday.

1 Introduction

In recent years, the malware industry has become a large and well-organized market [shahi2009]. Well funded, multi-player syndicates heavily invest in technologies and capabilities built to evade traditional protection, requiring anti-malware vendors to develop counter-mechanisms for finding and deactivating them. In the meantime, they inflict significant financial loss to users of computer systems.

One of the major challenges that anti-malware software faces today are the vast amounts of data which needs to be evaluated for potential malicious intent. For example, Microsoft’s real-time anti-malware detection products executes on over 600M computers worldwide [msrt18]. This generates tens of millions of daily data points to be analyzed as potential malware. One of the main reasons for these high volumes of different files is that in order to evade detection, malware authors introduce polymorphism to the malicious components. This means that malicious files belonging to the same malware “family”, with the same forms of malicious behavior, are constantly modified and/or obfuscated using various tactics, so that they appear to be many different files.

A first step in effectively analyzing and classifying such a large number of files is to group them and identify their respective families. In addition, such grouping criteria may be applied to new files encountered on computers in order to detect them as malicious and of a certain family. To facilitate research in this area, especially in the development of effective techniques for grouping variants of malware files into their respective families, Microsoft provided the data science and security communities with a malware dataset of unprecedented size. Here we summarize the many uses of this dataset, published to date.

2 Dataset

The malware dataset is almost half a terabyte when uncompressed. It consists of a set of known malware files representing a mix of 9 different families. Each malware file has an identifier, a 20 character hash value uniquely identifying the file, and a class label, which is an integer representing one of the 9 family names to which the malware may belong (See Table 1). For each file, the raw data contains the hexadecimal representation of the file’s binary content, without the header (to ensure sterility). The dataset also includes a metadata manifest, which is a log containing various metadata information extracted from the binary, such as function calls, strings, etc. This was generated using the IDA disassembler tool. The original question posed to participants was to classify malware to one of the 9 classes. The dataset can be downloaded from the competition website.222https://www.kaggle.com/c/malware-classification/data

Family Name # Train Samples Type
Ramnit 1541 Worm
Lollipop 2478 Adware
Kelihos_ver3 2942 Backdoor
Vundo 475 Trojan
Simda 42 Backdoor
Tracur 751 TrojanDownloader
Kelihos_ver1 398 Backdoor
Obfuscator.ACY 1228 Any kind of obfuscated malware
Gatak 1013 Backdoor
Table 1: Malware families in the dataset

3 Citations Comparison

Since the end of the competition in April 2015, more than 50 research papers and thesis works cited the competition and the dataset. Among the citations, several papers are not in English, which we are unable to read [russian2016, korean2016, french2015, NODE07303202]

. The remaining articles can be divided into two principal classes. The first category of papers referenced the challenge to either perform an abstract comparison or highlight the importance of machine learning for malware classification in industry, where the size of data is huge

[7413680, DBLP:journals/corr/GarciaM16, imran2016, DBLP:journals/corr/WangGQ16, DBLP:journals/corr/GaoWQ17, 8062202, ndss18, Scofield:2017:FML:3151137.3151142, 10.1007/978-981-10-7080-8_17, DBLP:journals/corr/abs-1708-08042, webeye2017, Hansen2017, 10.1007/978-3-319-60876-1_10, Oscar2017, Fields2016, Fowler2015, 8102915, 10.1007/978-3-319-73830-7_6]. Papers in the second category performed partial or complete evaluation on the dataset to verify the effectiveness and/or efficiency of their proposed approach for various tasks. We list the papers of the second category in Table 2

sorted by the publication date. Moreover, we summarize the main contribution or focus of each paper to make higher level clusters. Feature engineering, feature selection/fusion, being scalable, being robust, malware authorship attribution, detecting concept drift, performing a measurement, similarity hashing, classification techniques and deep learning are the major contributions of the papers. The diversity of the contributions has made the dataset a benchmark for various tasks, helping researchers provide a standard for evaluation and comparison.

Month Year Method Focus/Contribution of Research
Mar 2016 Ahmadi et al. [Ahmadi:2016:NFE:2857705.2857713] Feature Engineering,Feature Fusion,Being Scalable
May 2016 Drew et al. [7527757] Feature Engineering,Being Scalable
Jul 2016 Hu et al. [7523365] Being Scalable
Jul 2016 Narayanan et al. [7856826] Feature Engineering
Jul 2016 Celik et al. [DBLP:journals/corr/CelikMIPS16] Being Robust
Aug 2016 Zhang et al. [7847046] Being Scalable,Classification Techniques
Sep 2016 Bhattacharya et al. [DBLP:journals/corr/BhattacharyaMBC16] Being Scalable,Classification Techniques
Oct 2016 Dinh et al. [7723750] Classification Techniques
Oct 2016 Wojnowicz et al. [7796904] Feature Reduction
Nov 2016 Borbely [Borbely2016] Clustering Techniques
Dec 2016 Burnaev et al. [7836677] Classification Techniques
Dec 2016 Alrabaee et al. [10.1007/978-3-319-51966-1_17] Malware Authorship Attribution
Jan 2017 Drew et al. [Drew2017] Being Scalable,Classification Techniques
Jan 2017 Patri et al. [Patri2017DiscoveringMW] Classification Techniques
Mar 2017 Hassen et al. [Hassen:2017:SFC:3029806.3029824] Feature Engineering,Being Scalable
Mar 2017 Celik et al. [Celik:2017:FCP:3041008.3041018] Being Robust
May 2017 Yousefi-Azar et al. [7966342] Feature Engineering
Jun 2017 Kebede et al. [8268747] Deep Learning
Jul 2017 Yuxin et al. [Yuxin2017] Deep Learning
Aug 2017 Zhang et al. [8029425] Clustering Techniques
Aug 2017 Jordaney at al. [203684] Detecting Concept Drift
Aug 2017 Raff et al. [DBLP:journals/corr/abs-1708-03346] Similarity Hashing
Oct 2017 Kim et al. [Kim2017] Deep Learning
Nov 2017 Rahul et al. [10.1007/978-981-10-6898-0_19] Deep Learning
Dec 2017 Bagga [Bagga2017] Measurement and Comparison
Dec 2017 Gsponer et al. [10.1007/978-3-319-71246-8_3] Classification Techniques
Dec 2017 Hassen et al. [Hassen2017] Feature Engineering
Dec 2017 Fan et al. [8064195] Being Scalable
Dec 2017 Kim [10.1007/978-981-10-7605-3_215] Deep Learning
Jan 2018 Hwang et al. [fselection2018] Feature Selection
Feb 2018 Yan et al. [Yan2018scn] Deep Learning
Feb 2018 Kreuk et al. [2018arXiv180204528K] Adverserial Examples,Deep Learning
Feb 2018 Hassen et al. [2018arXiv180204365H] Open Set Recognition
Table 2: A comparison between research papers that have performed partial or complete evaluation on Microsoft malware classification challenge dataset.

4 Conclusion and Future Directions

In this paper, we provide a short description of the characteristics of the Microsoft Malware Classification Challenge dataset. This dataset is becoming a standard dataset with more than 50 papers citing it. We enumerated these references as much as possible and compared their main contributions with respect to the dataset. The comparison helps the understanding of what the existing contributions are, and what the potential research directions can be.

The authors aim to keep the reference table updated. We encourage the community to cite this paper when using the dataset, and update us about such work so it can be added to this paper.

References