UNSW_NB15_RNN
所属分类：其他
开发工具：Python
文件大小：27670KB
下载次数：4
上传日期：2020-11-25 20:13:19
上 传 者：
x潇研
说明： 用UNSW数据集进行入侵检测，运用各种组合模型，精确度能达到90%以上，运用比较流行的神经网络模型分别进行了测试
(Intrusion detection using UNSW dataset)
文件列表:
UNSW_NB15_RNN\build_model.py (3977, 2020-03-26)
UNSW_NB15_RNN\classifier.py (4486, 2020-03-26)
UNSW_NB15_RNN\data\encoded_test.npy (10538624, 2020-06-22)
UNSW_NB15_RNN\data\encoded_train.npy (22443776, 2020-06-22)
UNSW_NB15_RNN\data\test_label.npy (658784, 2020-06-22)
UNSW_NB15_RNN\data\train_label.npy (1402856, 2020-06-22)
UNSW_NB15_RNN\data\UNSW_NB15_testing-set.csv (15298467, 2018-04-29)
UNSW_NB15_RNN\data\UNSW_NB15_testing-set.rar (3503081, 2020-03-26)
UNSW_NB15_RNN\data\UNSW_NB15_training-set.csv (32117676, 2018-04-29)
UNSW_NB15_RNN\data\UNSW_NB15_training-set.rar (6926992, 2020-03-26)
UNSW_NB15_RNN\data_generator.py (5836, 2020-03-26)
UNSW_NB15_RNN\data_processing.py (1169, 2020-03-26)
UNSW_NB15_RNN\figure\framework.png (6162, 2020-03-26)
UNSW_NB15_RNN\figure\GRU.png (36441, 2020-03-26)
UNSW_NB15_RNN\figure\LSTM.png (42000, 2020-03-26)
UNSW_NB15_RNN\figure\Sparse AE.png (33601, 2020-03-26)
UNSW_NB15_RNN\figure\wave_1.png (778023, 2020-03-26)
UNSW_NB15_RNN\logs\events.out.tfevents.1592802476.MM-202005312146 (3104792, 2020-06-22)
UNSW_NB15_RNN\plot_wave_testing.py (2369, 2020-03-26)
UNSW_NB15_RNN\saved_ae_1\best_ae_1.hdf5 (637216, 2020-06-22)
UNSW_NB15_RNN\saved_ae_2\best_ae_2.hdf5 (128272, 2020-06-22)
UNSW_NB15_RNN\saved_ae_3\best_ae_3.hdf5 (53520, 2020-06-22)
UNSW_NB15_RNN\saved_models_temp\best_model.hdf5 (490520, 2020-06-22)
UNSW_NB15_RNN\__pycache__\build_model.cpython-35.pyc (3347, 2020-06-22)
UNSW_NB15_RNN\__pycache__\data_processing.cpython-35.pyc (1152, 2020-06-22)
UNSW_NB15_RNN\data (0, 2020-06-22)
UNSW_NB15_RNN\figure (0, 2020-03-26)
UNSW_NB15_RNN\logs (0, 2020-06-22)
UNSW_NB15_RNN\models (0, 2020-03-26)
UNSW_NB15_RNN\saved_ae_1 (0, 2020-06-22)
UNSW_NB15_RNN\saved_ae_2 (0, 2020-06-22)
UNSW_NB15_RNN\saved_ae_3 (0, 2020-06-22)
... ...
# Time-related-Intrusion-Detection-Model-based-on-Recurrent-Neural-Network
Here, we use RNN to deal with the network intrusion problem. The UNSW-NB15 dataset is used.
Totally, we divide the process into two parts.
The first part is regarded as pre-training. The stacked sparse autoencoder is used for feature extraction and dimension reduction.
The structure of SSAE is as follow:
After that, the data can be in low-dimension.
Then, we organize the 2D traffic into 3D data, that is, put a few samples together as a time-seires sample.
Finally, different variant RNNs are adopted to classify the current data into normal or anomaly.
The structure of LSTM is:
The structure of GRU is:
The final result of classification can be shown in this way:
If the method does help for you in your paper, please cite:
Lin Y, Wang J, Tu Y, et al. Time-Related Network Intrusion Detection Model: A Deep Learning Method[C]//2019 IEEE Global Communications Conference (GLOBECOM). IEEE, 2019: 1-6.
Likewise, if there is any question, contact me heuwangjie@hrbeu.edu.cn. Thanks.
本实验主要完成以下内容:
1.构建一个稀疏自编码器,完成降维任务
2.构建LSTM(GRU、双向LSTM、双向GRU),完成数据特征挖掘
3.最后以一个sigmoid神经元完成二分类任务,以binary_crossentropy作为衡量指标
预期:
1.不同DAE结构,对于分类的影响。
这个过程实际为pre-training过程,以DAE+单步LSTM(GRU及双向)结构完成任务
最终确定所需参数,包括DAE神经元个数以及稀疏系数rho的确定,需要给出一个表table
2.不同LSTM、GRU及双向,共4个的对比
这个过程实际为fine-tuning过程,以不同步数+结构完成任务
最终确定最优的结构,包括选用的cell是什么以及步数的确定,需要给出一个图figure(4条曲线,在不同步数的准确率和虚警率对比)
3.选出最优结构,然后进行test一维图的显示,figure
近期下载者：
相关文件：
收藏者：