DIVE: A Multi-Label Smart Contract Vulnerability Dataset

📄 Dataset Description

The DIVE dataset was constructed from raw data collected via the Etherscan API, using three main endpoints to capture:

• 🧩 Contract-level metadata

• 🧾 Account-level information

• ⚙️ Opcodes

DIVE integrates a wide variety of smart contract features, organized into two separate datasets:

• DIVE_PRE_Data (pre-deployment features)

• DIVE_POST_Data (post-deployment features)

Each dataset is available in two formats: Raw and Preprocessed. Following multiple preprocessing steps, the final processed versions provide:

• 22,330 smart contracts (samples)

• 221 features in DIVE_PRE_Data

• 176 features in DIVE_POST_Data

🏷️ Multi-Label Vulnerability Annotation

Since a single contract can exhibit multiple vulnerabilities, DIVE is structured as a multi-label dataset. Vulnerability labels were generated through the MultiTagging framework, which analyzes each contract’s source code using a suite of six established tools: MAIAN, Mythril, Semgrep, Slither, Solhint, and VeriSmart.  All tools were executed with consistent versions and configurations, aligned with the MultiTagging project specifications.

🛡️ Vulnerability Labels (DASP Top 10 Categories)

The dataset includes labels mapped to the first 8 categories of the DASP Top 10 vulnerability taxonomy:

1. Reentrancy

2. Access Control

3. Arithmetic Issues (Integer Overflow/Underflow)

4. Unchecked Call Return Values

5. Denial of Service (DoS)

6. Bad Randomness

7. Front Running

8. Time Manipulation (Timestamp Dependence)

🎯 Applications

The DIVE dataset is primarily intended for smart contract vulnerability detection through security analysis and machine learning.
Beyond this, its rich feature set and multi-label structure enable research across multiple domains, including:

• Vulnerability Detection

• Representation Learning

• Transfer & Domain Adaptation

• Feature Interpretability

• Anomaly Detection

📦 Package Contents

• DIVE_Raw_Data.zip – PRE- and POST-deployment unprocessed attributes.

• DIVE_Processed_Data.zip – PRE- and POST-deployment data ready for ML tasks.

• DIVE_Labels.zip –

  • DIVE_Labels.csv: Final multi-label vulnerability annotations.

  • Tool_Results.csv: Per-tool vulnerability flags mapped to DASP categories.

• Feature list.xlsx – Comprehensive feature catalog (name, type, description, category).

• EDA_and_Profiling_Reports.zip – Exploratory data analysis, profiling reports, and feature distribution plots for both PRE and POST datasets.