基于堆叠算法的代码混淆有效性评估模型  

Code obfuscation effectiveness evaluation model based on Stacking algorithm

作　　者：苏庆[1] 黄海滨 黄剑锋[1] 林佳锐 谢国波[1] SU Qing;HUANG Hai-bin;HUANG Jian-feng;LIN Jia-rui;XIE Guo-bo(School of Computer Science and Technology,Guangdong University of Technology,Guangzhou 510006,China)

机构地区：[1]广东工业大学计算机学院,广东广州510006

出　　处：《计算机工程与设计》2023年第3期755-761,共7页Computer Engineering and Design

基　　金：国家自然科学基金项目(618002072);广东省自然科学基金项目(2018A030313389);广东省高等教育教学改革基金项目(SJJG20191216)。

摘　　要：为解决代码混淆算法有效性评估模型存在评价指标不全面、单一学习器泛化能力低的问题,提出一种融合自适应增强训练机制和Stacking算法的代码混淆算法有效性评估模型SDF-Stacking。构建一个包括强度、弹性、开销、隐蔽性4大特征的代码混淆有效性评价指标集合;在模型的基分类器训练阶段引入自适应增强训练机制,提高基分类器的预测精度和多样性;使用最大互信息算法做数据融合,增大元分类器训练数据信息量。实验结果表明,该模型在多个评价指标上均优于其它对比模型,准确率可达98.6%。To solve the problems of incomplete evaluation indexes and low generalization ability of a single learner in the code obfuscation algorithm effectiveness evaluation model,a code obfuscation algorithm effectiveness evaluation model SDF-Stacking was proposed,which integrated adaptive enhancement training mechanism and Stacking algorithm.A set of code obfuscation effectiveness evaluation indexes including four characteristics of strength,resilience,overhead,and concealment was constructed.An adaptive enhancement training mechanism was introduced in the base classifier training stage to improve the prediction accuracy and diversity of the base classifier.A maximum mutual information algorithm was used to do data fusion to increase the amount of information in the meta classifier training data.Experimental results show that the proposed model outperforms other compa-rison models in several evaluation metrics,and the accuracy rate can reach 98.6%.

关 键 词：Stacking算法 自适应增强训练 代码混淆 数据采样权重 混淆算法有效性评估 特征提取 特征选择 

分 类 号：TP311[自动化与计算机技术—计算机软件与理论]