IDPonzi: An interpretable detection model for identifying smart Ponzi schemes

Ponzi schemes are deceptive financial scams that lure users with the promise of high profits, resulting in substantial losses for global investors. The advent of blockchain technologies has witnessed these traditional scams transitioning from offline operations to the blockchain system. In the blockchain environment, Ponzi schemes often take the form of high-return investment contracts. Existing approaches for detecting smart Ponzi schemes rely on machine learning techniques that analyze smart contracts’ operation codes or transaction histories. However, these approaches, which rely on the frequency distribution of opcodes, lack interpretability. Additionally, transaction-based methods require a significant number of generated transactions, limiting their ability to promptly detect newly deployed smart contracts. These limitations render current detection methods inefficient in identifying Ponzi schemes. This paper proposes IDPonzi, a novel interpretable model for detecting smart Ponzi schemes in the blockchain. We refine the dataset by eliminating duplicate contracts to enhance the detection capability, resulting in more compact and discriminative samples. We then utilize a classification algorithm to analyze the features extracted from the operation codes of contracts, accurately identifying Ponzi schemes. Specifically, we employ the Shapley Additive exPlanation (SHAP) method to interpret predictions for individual samples and conduct a dependency analysis for four pairs of features. Experimental results demonstrate that IDPonzi achieves impressive effectiveness, with a precision of 99%, recall of 85%, and F-score of 92%, outperforming existing approaches. •A novel detection model that identifies smart Ponzi schemes (a kind of financial scams) from smart contracts in Blockchain.•Interpret the complex non-linear results underlying smart contracts using Shapley Additive explanation.•Achieve 99% precision, 85% recall, and 92% F-score.