This code is a implementation of Web-Application-Firewall driven by deep learning model, a nice work on AI-WAF.
- Dataset
- good and bad queries
- Usage
- Training
- Example
- Demo
- Reference
This repository contains:
- model code which implemented the paper.
- data loader code you can use to load dataset for training data.
- training scripts to train the model.
pip install -r requirements.txtpython train.py -o 1 -b 8 -e 30a trained model which use private dataset, stored in ./cache dir
__________________________________________________________________________________________________
Layer (type) Output Shape Param # Connected to
==================================================================================================
input_1 (InputLayer) [(None, 1024)] 0
__________________________________________________________________________________________________
embedding (Embedding) (None, 1024, 20) 1229180 input_1[0][0]
__________________________________________________________________________________________________
spatial_dropout1d (SpatialDropo (None, 1024, 20) 0 embedding[0][0]
__________________________________________________________________________________________________
conv1d (Conv1D) (None, 1024, 32) 672 spatial_dropout1d[0][0]
__________________________________________________________________________________________________
conv1d_1 (Conv1D) (None, 1024, 32) 1952 spatial_dropout1d[0][0]
__________________________________________________________________________________________________
conv1d_2 (Conv1D) (None, 1024, 32) 3232 spatial_dropout1d[0][0]
__________________________________________________________________________________________________
max_pooling1d (MaxPooling1D) (None, 512, 32) 0 conv1d[0][0]
__________________________________________________________________________________________________
max_pooling1d_1 (MaxPooling1D) (None, 512, 32) 0 conv1d_1[0][0]
__________________________________________________________________________________________________
max_pooling1d_2 (MaxPooling1D) (None, 512, 32) 0 conv1d_2[0][0]
__________________________________________________________________________________________________
concatenate (Concatenate) (None, 512, 96) 0 max_pooling1d[0][0]
max_pooling1d_1[0][0]
max_pooling1d_2[0][0]
__________________________________________________________________________________________________
flatten (Flatten) (None, 49152) 0 concatenate[0][0]
__________________________________________________________________________________________________
dropout (Dropout) (None, 49152) 0 flatten[0][0]
__________________________________________________________________________________________________
dense (Dense) (None, 256) 12583168 dropout[0][0]
__________________________________________________________________________________________________
dropout_1 (Dropout) (None, 256) 0 dense[0][0]
__________________________________________________________________________________________________
dense_1 (Dense) (None, 2) 514 dropout_1[0][0]
==================================================================================================
Total params: 13,818,718
Trainable params: 13,818,718
Non-trainable params: 0
Restoring model weights from the end of the best epoch.
Accuracy Score is: 0.9809785480291662
Precision Score is : 0.988558352402746
Recall Score is : 0.9707865168539326
F1 Score: 0.9795918367346939
AUC Score: 0.9804062247146184
__________________________________________________________________________________________________
Layer (type) Output Shape Param # Connected to
==================================================================================================
input_1 (InputLayer) [(None, 1024)] 0
__________________________________________________________________________________________________
embedding (Embedding) (None, 1024, 128) 7866368 input_1[0][0]
__________________________________________________________________________________________________
spatial_dropout1d (SpatialDropo (None, 1024, 128) 0 embedding[0][0]
__________________________________________________________________________________________________
conv1d (Conv1D) (None, 1022, 8) 3080 spatial_dropout1d[0][0]
__________________________________________________________________________________________________
conv1d_1 (Conv1D) (None, 1020, 8) 5128 spatial_dropout1d[0][0]
__________________________________________________________________________________________________
conv1d_2 (Conv1D) (None, 1018, 8) 7176 spatial_dropout1d[0][0]
__________________________________________________________________________________________________
conv1d_3 (Conv1D) (None, 1016, 8) 9224 spatial_dropout1d[0][0]
__________________________________________________________________________________________________
conv1d_4 (Conv1D) (None, 1014, 8) 11272 spatial_dropout1d[0][0]
__________________________________________________________________________________________________
global_average_pooling1d (Globa (None, 8) 0 conv1d[0][0]
__________________________________________________________________________________________________
global_max_pooling1d (GlobalMax (None, 8) 0 conv1d[0][0]
__________________________________________________________________________________________________
global_average_pooling1d_1 (Glo (None, 8) 0 conv1d_1[0][0]
__________________________________________________________________________________________________
global_max_pooling1d_1 (GlobalM (None, 8) 0 conv1d_1[0][0]
__________________________________________________________________________________________________
global_average_pooling1d_2 (Glo (None, 8) 0 conv1d_2[0][0]
__________________________________________________________________________________________________
global_max_pooling1d_2 (GlobalM (None, 8) 0 conv1d_2[0][0]
__________________________________________________________________________________________________
global_average_pooling1d_3 (Glo (None, 8) 0 conv1d_3[0][0]
__________________________________________________________________________________________________
global_max_pooling1d_3 (GlobalM (None, 8) 0 conv1d_3[0][0]
__________________________________________________________________________________________________
global_average_pooling1d_4 (Glo (None, 8) 0 conv1d_4[0][0]
__________________________________________________________________________________________________
global_max_pooling1d_4 (GlobalM (None, 8) 0 conv1d_4[0][0]
__________________________________________________________________________________________________
concatenate (Concatenate) (None, 80) 0 global_average_pooling1d[0][0]
global_max_pooling1d[0][0]
global_average_pooling1d_1[0][0]
global_max_pooling1d_1[0][0]
global_average_pooling1d_2[0][0]
global_max_pooling1d_2[0][0]
global_average_pooling1d_3[0][0]
global_max_pooling1d_3[0][0]
global_average_pooling1d_4[0][0]
global_max_pooling1d_4[0][0]
__________________________________________________________________________________________________
flatten (Flatten) (None, 80) 0 concatenate[0][0]
__________________________________________________________________________________________________
dropout (Dropout) (None, 80) 0 flatten[0][0]
__________________________________________________________________________________________________
dense (Dense) (None, 256) 20736 dropout[0][0]
__________________________________________________________________________________________________
dropout_1 (Dropout) (None, 256) 0 dense[0][0]
__________________________________________________________________________________________________
dense_1 (Dense) (None, 2) 514 dropout_1[0][0]
==================================================================================================
Total params: 7,923,498
Trainable params: 7,923,498
Non-trainable params: 0
Restoring model weights from the end of the best epoch.
Accuracy Score is: 0.978862819699852
Precision Score is : 0.9851718714895529
Recall Score is : 0.9703474219960169
F1 Score: 0.9777034559643255
AUC Score: 0.9784976033710613
__________________________________________________________________________________________________
Layer (type) Output Shape Param # Connected to
==================================================================================================
input_1 (InputLayer) [(None, 1024)] 0
__________________________________________________________________________________________________
embedding (Embedding) (None, 1024, 128) 7851520 input_1[0][0]
__________________________________________________________________________________________________
conv1d (Conv1D) (None, 1024, 512) 131584 embedding[0][0]
__________________________________________________________________________________________________
conv1d_1 (Conv1D) (None, 1024, 512) 197120 embedding[0][0]
__________________________________________________________________________________________________
conv1d_2 (Conv1D) (None, 1024, 512) 262656 embedding[0][0]
__________________________________________________________________________________________________
concatenate (Concatenate) (None, 3072, 512) 0 conv1d[0][0]
conv1d_1[0][0]
conv1d_2[0][0]
__________________________________________________________________________________________________
flatten (Flatten) (None, 1572864) 0 concatenate[0][0]
__________________________________________________________________________________________________
dropout (Dropout) (None, 1572864) 0 flatten[0][0]
__________________________________________________________________________________________________
dense (Dense) (None, 2) 3145730 dropout[0][0]
==================================================================================================
Total params: 11,588,610
Trainable params: 11,588,610
Non-trainable params: 0
Restoring model weights from the end of the best epoch.
Accuracy Score is: 0.9844674556213018
Precision Score is : 0.9935260115606936
Recall Score is : 0.973052536231884
F1 Score: 0.9831827022079854
AUC Score: 0.9837528925216473
Samples:
If this project help you reduce time to develop, you can give me a cup of coffee :)
AliPay(支付宝)
WechatPay(微信)
MIT © Kun