hendriow/mh100k

MH-100K: An innovative Android Malware Dataset for advanced research

Dataset Summary

MH-100K is a large-scale dataset for Android malware detection research. It contains 101,975 Android applications (APKs) collected between 2010 and 2022, providing a diverse set of samples to study malware evolution over more than a decade.

The dataset features high-dimensional tabular data representing the static analysis of these applications. It includes permissions, API calls, and intents, along with extensive metadata and detection labels derived from VirusTotal.

Dataset Structure

The repository contains the dataset in a consolidated, efficient format:

• mh100.parquet: The main dataset file containing the feature matrix and metadata for all 101,975 instances. Stored in int8 format for efficiency.
• mh100-labels.csv: Contains the label information (Malware vs Benign) and VirusTotal metadata.
• feature_names.csv: A mapping file that lists the names of the features corresponding to the columns in the feature matrix.

How to Use

You can load this dataset directly using the Hugging Face datasets library.

Quick Load

from huggingface_hub import hf_hub_download
import pandas as pd

# 1. Download the specific file to your cache
file_path = hf_hub_download(
    repo_id="hendriow/mh100k",
    filename="mh100.parquet",
    repo_type="dataset"
)

# 2. Read it directly into a dataframe
df = pd.read_parquet(file_path)

df.info()

Loading with Feature Names

Since the dataset is high-dimensional (>10k features), the columns in the parquet file might be indexed. You can map them back to their real names (e.g., android.permission.INTERNET) using the feature_names.csv file.

from huggingface_hub import hf_hub_download
import pandas as pd

# 1. Download the labels file to your local cache
csv_path = hf_hub_download(
    repo_id="hendriow/mh100k",
    filename="mh100-labels.csv",
    repo_type="dataset"
)

# 2. Read into a DataFrame
labels_df = pd.read_csv(csv_path)

labels_df.head()

Dataset Description

The MH-100K dataset is a comprehensive collection of Android malware information, comprising 101,975 samples.

• Data Type: Tabular (Int8)
• Time Period: 2010 - 2022
• Source: Samples randomly selected from AndroZoo.

Features and Attributes

• SHA256 hash (APK's signature)
• File name
• Package name
• Android's official compilation API
• 166 permissions
• 24,417 API calls
• 250 intents

About VirusTotal API

The VirusTotal API is a crucial tool in this dataset's creation, known for its prowess in detecting suspicious files and URLs. Each API request yields a JSON, aiding in categorizing the APK based on its perceived threat.

Citation

If you use this dataset in your research, please cite the original authors:

@article{bragancca2023android, title={Android malware detection with MH-100K: An innovative dataset for advanced research}, author={Bragan{\c{c}}a, Hendrio and Rocha, Vanderson and Barcellos, Lucas and Souto, Eduardo and Kreutz, Diego and Feitosa, Eduardo}, journal={Data in Brief}, volume={51}, pages={109750}, year={2023}, publisher={Elsevier} }

@inproceedings{bragancca2023capturing,
title={Capturing the behavior of android malware with mh-100k: A novel and multidimensional dataset},
author={Bragan{\c{c}}a, Hendrio and Rocha, Vanderson and Barcellos, Lucas Vilanova and Souto, Eduardo and Kreutz, Diego and Feitosa, Eduardo},
booktitle={Simp{'o}sio Brasileiro de Seguran{\c{c}}a da Informa{\c{c}}{~a}o e de Sistemas Computacionais (SBSeg)},
pages={510--515},
year={2023},
organization={SBC} }