In this paper, a cascaded depth learning framework is constructed. A cascaded depth model is successfully implemented by studying and analyzing the specific feature transformation, feature selection, and classifier algorithm used in the framework. A feature combination method based on enhanced feature selection and classification is proposed according to the different features learned by each layer of the model. Combining block chain cryptography technology, distributed technology, consensus accounting mechanism of technology innovation, transaction data encapsulation into specific format data unit, encapsulated into a linear list in chronological order, using encryption algorithm trading transparency, traceability of data, security, credibility, and uniqueness in financial data analysis. The experimental results show that with the increase of the number of model layers, our method can significantly improve the classification accuracy. This result also verifies that the proposed model can learn more effective data representation features and also verifies the effectiveness of the proposed feature combination method.

本文构建了一个级联的深度学习框架。通过研究和分析框架中使用的特定特征转换，特征选择和分类器算法，成功实现了级联深度模型。根据模型各层学习到的不同特征，提出了一种基于增强特征选择和分类的特征组合方法。结合区块链密码技术，分布式技术，技术创新的共识会计机制，将交易数据封装成特定格式的数据单元，按时间顺序封装成线性列表，使用加密算法实现交易透明性，数据的可追溯性，安全性，可信性和唯一性在财务数据分析中。实验结果表明，随着模型层数的增加，我们的方法可以显着提高分类精度。该结果还验证了所提出的模型可以学习更多有效的数据表示特征，还验证了所提出的特征组合方法的有效性。