Paper: arXiv 2308.16391

Authors: Phuong Duy Huynh, Son Hoang Dau, Xiaodong Li, Phuc Luong, Emanuele Viterbo

Abstract

The Ponzi scheme, an old-fashioned fraud, is now popular on the Ethereum blockchain, causing considerable financial losses to many crypto investors. A few Ponzi detection methods have been proposed in the literature, most of which detect a Ponzi scheme based on its smart contract source code. This contract-code-based approach, while achieving very high accuracy, is not robust because a Ponzi developer can fool a detection model by obfuscating the opcode or inventing a new profit distribution logic that cannot be detected. On the contrary, a transaction-based approach could improve the robustness of detection because transactions, unlike smart contracts, are harder to be manipulated. However, the current transaction-based detection models achieve fairly low accuracy. In this paper, we aim to improve the accuracy of the transaction-based models by employing time-series features, which turn out to be crucial in capturing the life-time behaviour a Ponzi application but were completely overlooked in previous works. We propose a new set of 85 features (22 known account-based and 63 new time-series features), which allows off-the-shelf machine learning algorithms to achieve up to 30% higher F1-scores compared to existing works.

Complexity vs Empirical Score

  • Math Complexity: 3.0/10
  • Empirical Rigor: 7.5/10
  • Quadrant: Street Traders — empirical and applied, lower math complexity

Why this score: The paper relies on standard machine learning algorithms and feature engineering rather than advanced mathematical derivations, but demonstrates high empirical rigor through extensive experimentation with real blockchain transaction data, feature selection, and comparison against multiple baselines.

Research Flowchart

  flowchart TD
  A["Research Goal:<br>Improve accuracy of<br>transaction-based Ponzi detection"] --> B{"Methodology"};
  B --> C["Feature Engineering<br>85 total features<br>• 22 known account-based<br>• 63 new time-series"];
  B --> D["Dataset<br>Ethereum Transactions"];
  C --> E["Model Training<br>Off-the-shelf ML Algorithms"];
  D --> E;
  E --> F["Model Evaluation<br>F1-score metric"];
  F --> G["Key Outcome<br>Up to 30% higher F1-scores<br>vs. existing works"];

Evidence Card

What the authors claim, on what data, and who has checked it. Read from the paper's own text by a language model: every quote is verbatim, nothing was re-run. How to read this card

Proposes

The paper proposes a new set of 85 features (22 known account-based and 63 new time-series features) for transaction-based Ponzi detection on Ethereum.

Source sentence
We propose a new set of 85 features (22 known account-based and 63 new time-series features), which allows off-the-shelf machine learning algorithms to achieve up to 30% higher F1-scores compared to existing works.

Assumes 2

  • Transactions are harder to manipulate than smart contract code, making a transaction‑based approach more robust. data
    Source sentence
    a transaction-based approach could improve the robustness of detection because transactions, unlike smart contracts, are harder to be manipulated
  • Applications with no transactions or lifetime shorter than one day are outliers and should be removed to build a clean dataset. data
    Source sentence
    We also discarded applications with no transactions or having lifetime (the period from the recreation time to the last transaction) shorter than one day. These are outliers, whose behaviours do not resemble the whole group.

Data

  • Refined Ponzi and non‑Ponzi address dataset · XBlock‑ETH repository (based on SADPonzi paper) crypto-onchain public
    Source sentence
    We refined the dataset of labeled Ponzi and non-Ponzi addresses provided in the SADPonzi paper [14] by first downloading and extracting transaction histories

Costs

Not applicable The paper reports no trading or execution result.

Finds 4

  • The new 85‑feature set enables up to 30% higher F1‑scores compared to existing transaction‑based works. predictive-accuracy
    Source sentence
    features ... which allows off-the-shelf machine learning algorithms to achieve up to 30% higher F1-scores compared to existing works.
  • Time‑series features contribute up to a 5.7% improvement in F1‑score over models using only account features. predictive-accuracy
    Source sentence
    The improvement is up to 5.7% as compared to models that only used account features.
  • The trimmed 85‑feature set outperforms the original 545‑feature set on all performance metrics across five machine‑learning algorithms. performance
    Source sentence
    It is remarkable that the trimmed feature set outperforms the original one for all performance metrics (Accuracy, Precision, Recall, F1-score, and running time) in all five off-the-shell machine learning algorithms.
  • The proposed model achieves 11% higher F1‑score than the Random Forest baseline [32] and 30% higher than the XGBoost baseline [12]. predictive-accuracy
    Source sentence
    our model achieved 11% higher F1-score compared to [32] (Random Forest) and 30% higher F1-score compared to [12] (XGBoost).

Checked by others

No other paper in the archive has been found to support or contradict these findings (compared with 22 related papers). Coverage is partial, so this is neither replication nor refutation.

Code

No repository by the authors found in the paper.

Extracted 2026-11-14 · same card as JSON · A language model chose these items: check the source sentence before relying on one.