基于多阶融合训练的恶意程序检测

Malware Detection Based on Multi-level Fusion Training

当前互联网上恶意软件极度泛滥,这些恶意软件不仅影响用户正常使用计算机,甚至还会造成破坏和经济损失,由恶意软件造成的损失事件频发,急需具有良好效果的检测模型。当前针对恶意软件检测的研究成果主要集中在特征工程和算法选择上,由于恶意软件反检测能力的进步,传统的检测方法已经不能应对具备对抗检测能力的恶意软件。为了解决此类问题,从特征融合和模型融合两个角度,用多阶训练的思想,建立了一种更有效的恶意软件检测模型。在真实数据集上的评估结果表明,该模型具有优秀的检测效果,准确率等相关评估指标均达到0.97以上的水平。

Malware on the Internet is currently extremely widespread,these malware not only affects the normal use of com⁃puters,and even cause damage and economic losses,the loss caused by malware is frequent,so there is an urgent need for detec⁃tion models with good results.Current research results on malware detection mainly focus on feature engineering and algorithm se⁃lection.Due to the development of malware counter-detection capability,traditional detection methods can no longer cope with mal⁃ware with counter-detection capability.In order to solve such problems,a more effective malware detection model is established from two perspectives of feature fusion and model fusion,using the idea of multi-level training.The evaluation results on a real da⁃taset show that the model has excellent detection effect,and the accuracy and other related evaluation indexes reach the level of 0.97 or above.