基于协同采样主动学习的恶意代码检测

Malware detection using active learning based on collaborative sampling

研究了基于机器学习分类算法的恶意代码检测,考虑到目前主要采用传统分类方法对恶意代码进行分类识别,这些方法需要通过学习大量标记样本来获得精准的分类器模型,然而样本标记工作只有少数专家才能完成,导致标记样本往往不足,致使分类结果准确率不高,提出了一种基于协同采样的主动学习方法。运用这种学习方法,仅需少量标记样本即可有效识别出恶意代码。实验证明,相对于传统的恶意代码分类方法,该方法能够显著提升分类准确率和泛化性能。

The malware detection using classification algorithms based on machine learning was studied. In consideration of the fact that current malware recognition mainly uses traditional classification algorithms, thus leading to the application of machine learning models and low classification precision due to the unsufficiency of labelled samples, a new malware detection method using active learning based on collaborative sampling was proposed. The method can use less labelled samples to effectively recognize malware. The experiment showed that it had the higher classifica- tion precision and the better performance compared with traditional methods.