Abstract: Aiming at the poor detection performance of traditional intrusion detection models in high-dimensional data and data imbalance environment, an intrusion detection model combining Adaptive Synthetic Sampling Approach（ADASYN） with improved Stacked Denoising Autoencoder（SDA） is proposed. Firstly, the data oversampling process is performed using the ADASYN approach. Secondly, the Adam optimization approach and Dropout regularization are used to improve the SDA deep learning model, and the integration features of low dimensionality and high robustness are extracted. Finally, intrusion detection is performed in the softmax classifier. Experimental results show that compared with SDA, AE-DNN and MSVM, the ADASYN-SDA model has a certain degree of improvements in average accuracy, detection rate and false positive rate.

Key words: Stacked Denoise Autoencoder（SDA）, Adaptive Synthetic Sampling Approach（ADASYN）, deep learning, intrusion detection

摘要： 针对传统入侵检测模型在高维数据且数据不均衡环境下检测性能较差的问题，提出了一种自适应过采样算法（ADASYN）与改进堆叠式降噪自编码器（SDA）结合的入侵检测模型。使用ADASYN算法进行数据过采样处理。使用Adam优化算法，以及Dropout正则化对SDA深度学习模型进行改进，提取出低维数、高鲁棒性的集成特征。在softmax分类器中进行入侵检测识别。实验结果表明，ADASYN-SDA模型相较于SDA、AE-DNN和MSVM模型，在平均准确率、检测率和误判率上均有一定程度的提高。

关键词: 堆叠式降噪自编码器（SDA）, 自适应过采样算法（ADASYN）, 深度学习, 入侵检测