基于交易特征对以太网多类型非法账户的分析与预测

日益频繁的非法交易行为妨害以太坊安全交易，针对电子货币的匿名性使得非法交易行为难于跟踪分析问题。以太坊平台交易数据作为数据源，以被标记得非法账户和未标记的正常账户数据集作为训练集，利用交易数据的特征属性为构造基础，通过CatBoost算法对其中包含多种类型的非法账户进行整体预测。其过程通过T-SNE算法实现交易特征的降维可视化，采用多倍交叉验证，引入SHAP value因子判断特征影响的正负属性，所建立模型的预测效果准确率达到了94.29%，感受者曲线下面积（AUC）数值的评估度量达到了0.984 6。建议的方案能较为准确地预测以太坊交易平台上存在的非法行为，有效改善基于区块链的交易环境。

Analysis and prediction of multi-type illegal accounts of ether based on transaction characteristics

The increasingly frequent illegal transactions hinder the secure transactions of Ethereum, and the anonymity of electronic currency makes it difficult to track and analyze illegal transactions. This paper used the transaction data of the Ethereum platform as the data source, the marked illegal account and unmarked normal account data set as the training set, and the characteristic attributes of the transaction data as the construction basis. Account for the overall forecast, the process used the T-SNE algorithm to realize the dimensionality reduction and visualization of transaction features, adopted multiple cross-validation, and introduced the SHAP Value factor to judge the positive and negative attributes of the feature. The prediction effect accuracy rate of the established model reached 94.29%. The evaluation metric for the area(AUC) value reached 0.984 6. The proposed scheme can more accurately predict the illegal behavior on the Ethereum trading platform, and will effectively improve the blockchain-based trading environment.