---
license: cc-by-4.0
task_categories:
  - tabular-classification
language:
  - en
tags:
  - malware-detection
  - android
  - multimodal
  - concept-drift
  - cybersecurity
  - benchmark
  - longitudinal
  - graph-neural-network
  - static-analysis
  - Croissant
pretty_name: "McNdroid"
size_categories:
  - 100K<n<1M
---

# McNdroid: A Longitudinal Multimodal Benchmark for Robust Drift Detection in Android Malware


## Dataset Description

McNdroid is a large-scale, longitudinal, multimodal Android malware detection dataset designed to benchmark concept drift robustness. It spans samples collected from 2013 to 2025 and provides three complementary modalities: static feature vectors, API call graphs (GML), and JSON-based behavioral representations. The dataset also includes a rich metadata CSV and per-vendor family-level verdicts, supporting fine-grained label analysis and multi-label learning.

### Dataset Summary

- **Modalities:** Static features (NPZ), API call graphs (GML), JSON behavioral features
- **Time span:** 2013–2025
- **Total size:** ~10.9 GB
- **Splits:** Train/test per year with temporal evaluation protocols
- **Labels:** Binary (malware/benign) and multi-vendor family-level verdicts


### Supported Tasks

- Android malware detection (binary classification)
- Concept drift detection and temporal robustness evaluation
- Multi-modal learning for malware analysis
- Graph-based malware classification


## Dataset Structure

### Repository Layout

```
McNdroid/
├── README.md
├── metadata.csv                         # Sample-level metadata (~87 MB)
├── vendor_family_wide_verdict.csv       # Multi-vendor family verdicts (~770 MB)
├── data_feature/                        # Static feature modality
│   └── processed_data/
│       └── init_2013/
│           ├── 2013/
│           │   ├── train_X.npz          # Training feature matrix (sparse)
│           │   ├── test_X.npz           # Test feature matrix (sparse)
│           │   ├── train_meta.npz       # Training labels and metadata
│           │   ├── test_meta.npz        # Test labels and metadata
│           │   ├── vocab.json           # Feature vocabulary
│           │   ├── selector_meta.json   # Feature selector metadata
│           │   └── split_meta.json      # Split statistics
│           ├── 2014/
│           ├── ...
│           └── 2025/
├── gml_feature/                         # API call graph modality (GML files)
│   └── processed_data/
│       └── ...
├── json_feature/                        # JSON behavioral feature modality
│   └── processed_data/
│       └── ...
```

### Data Fields

#### metadata.csv

Contains per-sample metadata including SHA256 hashes, collection timestamps, labels, and source information.

#### vendor_family_wide_verdict.csv

Contains malware family labels from multiple antivirus vendors, enabling multi-label and label-noise research.

#### Static Features (data_feature/)

Each year folder contains:
- `train_X.npz` / `test_X.npz`: Sparse feature matrices in NumPy compressed format
- `train_meta.npz` / `test_meta.npz`: Associated labels and sample metadata
- `vocab.json`: Feature name vocabulary mapping
- `selector_meta.json`: Feature selection metadata
- `split_meta.json`: Train/test split statistics

#### Graph Features (gml_feature/)

API call graphs stored in GML format, organized by year. Each graph represents inter-procedural API call relationships extracted via static analysis.

#### JSON Features (json_feature/)

Behavioral feature representations stored as JSON files, organized by year.


## Dataset Creation

### Source Data

Samples were collected from public malware repositories and benign application stores spanning 2013–2025. Each sample was processed through a static analysis pipeline to extract permissions, API calls, intents, and other manifest and bytecode-level features.

### Annotations

Labels are derived from VirusTotal multi-scanner verdicts. The `vendor_family_wide_verdict.csv` file preserves per-vendor family attributions to support research on label noise and disagreement.


## Considerations for Using the Data


### Social Impact

This dataset is intended for defensive cybersecurity research. It should not be used to develop offensive malware capabilities.

### Licensing

This dataset is released under the [Creative Commons Attribution 4.0 International License (CC-BY-4.0)](https://creativecommons.org/licenses/by/4.0/).


## Usage

### Loading Static Features

```python
import numpy as np
import json

# Load a specific year's training data
train_X = np.load("data_feature/processed_data/init_2013/2013/train_X.npz", allow_pickle=True)
train_meta = np.load("data_feature/processed_data/init_2013/2013/train_meta.npz", allow_pickle=True)

with open("data_feature/processed_data/init_2013/2013/vocab.json") as f:
    vocab = json.load(f)
```

### Loading Metadata

```python
import pandas as pd

metadata = pd.read_csv("metadata.csv")
verdicts = pd.read_csv("vendor_family_wide_verdict.csv")
```


## Citation

If you use this dataset, please cite:

```bibtex
[More Information Needed]
```

## Contact

For questions or issues, please open a discussion on the [Community tab](https://huggingface.co/datasets/IQSeC-Lab/McNdroid/discussions).