Build a Learning Model to detect IOT malware Detection

Task 3: CDMC2021 IoT Malware Detection

Based on the control flow graphs (CFGs) generated by a static-analysis tool, Radare2, and labels that indicating whether the samples are malware programs, the participants are required to perform an IoT malware detection task to predict whether the samples in the test set are malware or not. The dataset consists of 54,829 samples, which are generated from the following procedure: (1) a collection of malicious and benign Linux programs in ELF format were collected from various sources; (2) each of these programs are fed to Radare2 to extract the CFG information; and (3) JSON output from Radare2 that can be interpreted as a list of directed-graph components are then reformulate as a single line in a text file. Please see the “File Format” section for more detail.

Label (1: malware, 0: benign ware) of the ELF files are determined by the state-of-art anti-virus engines.

List of Files

The [login to view URL] file contains feature information of 16,521files in the training set.

The [login to view URL] file contains label information of 16,521files in the training set.

The [login to view URL] file contains information of 38,550 files in the testing set.

File Format

Steps to formulate the features.

Radare2 outputs its analysis result for an ELF sample program as a JSON object looks like the following.

[{"name": "sym.__uClibc_main", "imports": ["[login to view URL]", "sym.__GI_memcpy", "sym._dl_aux_init", "sym.__uClibc_init"]}, {"name": "sym._fp_out_narrow", "imports": ["sym.__GI_strlen ", "sym._charpad", "sym.__stdio_fwrite"]}, …]

Then, each node in the list is represented as a list of function calls with the “name” field placed at first, followed by the function calls in the “import” field. The components in the list are separated by white spaces. The JSON object above is changed to a list of nodes as follows.

Node 1: "sym.__uClibc_main" "[login to view URL]" "sym.__GI_memcpy" "sym._dl_aux_init" "sym.__uClibc_init"

Node 2: "sym._fp_out_narrow" "sym.__GI_strlen" "sym._charpad" "sym.__stdio_fwrite"

Nodes 3~: …

All nodes in the JSON list are sequentially joined by semicolons to form a single line in a .data file. Now, each line in the .data file corresponds to a single file in the dataset.

Line 1: "sym.__uClibc_main" "[login to view URL]" "sym.__GI_memcpy" "sym._dl_aux_init" "sym.__uClibc_init";"sym._fp_out_narrow" "sym.__GI_strlen" "sym._charpad" "sym.__stdio_fwrite";…

Task

The participants are required to provide the prediction of labels of the test samples based on information provided in the task.

Machine Learning (ML) Deep Learning Linguagem Natural Processamento de dados

ID do Projeto: #31566368

Sobre o projeto