Method and apparatus for detecting anomaly traffic
Disclosed is a method for detecting anomaly traffic using an autoencoder based on an inception-residual network-v2 which is one type of machine learning methods, the method including generating a plurality of encoders on the basis of traffic data for learning; generating a plurality of pieces of image data on the basis of traffic data for each session, which is a detection target; and determining whether the traffic data for each session is abnormal using at least one among the traffic data for learning, the plurality of encoders, and the plurality of pieces of image data.
1. A method for detecting anomaly traffic, comprising:
generating a plurality of different encoders on the basis of traffic data for learning;
generating a plurality of pieces of image data on the basis of traffic data for each session, which is a detection target;
determining whether the traffic data for each session is abnormal based on binary cross entropy (BCE) of the plurality of pieces of image data and a preset threshold value;
generating data clusters according to each of the plurality of different encoders using a clustering algorithm based on the traffic data for learning and the plurality of different encoders when the traffic data for each session is determined normal based on the BCE;
generating output data by inputting the plurality of image data to each of the plurality of different encoders for each image data sequentially; and
re-determining whether the traffic data for each session is abnormal based on whether the output data included in the data cluster according to each of the different encoders;
wherein the plurality of different encoders outputs different data for one image data, and wherein the generating a plurality of pieces of image data comprises:
converting each character constituting a character string included in the traffic data for each session into a plurality of one-hot vectors in a reverse order;
generating a matrix by combining the plurality of one-hot vectors; and
generating an image representing the location of each character in the matrix.
2. The method of claim 1 , wherein the generating of the plurality of different encoders on the basis of traffic data for learning includes:
dividing the traffic data for learning into a preset unit;
converting the divided traffic data for learning into a plurality of pieces of image data for learning; and
generating the plurality of different encoders on the basis of the plurality of pieces of image data for learning.
3. The method of claim 2 , wherein the generating of the plurality of different encoders on the basis of the plurality of pieces of image data for learning includes learning the plurality of pieces of image data for learning through an autoencoder based on an inception-residual network-v2 using convolutional neural networks to generate the plurality of different encoders.
4. The method of claim 3 , wherein the determining whether the traffic data for each session is abnormal based on BCE of the plurality of pieces of image data and a preset threshold value includes:
calculating BCE by inputting the plurality of pieces of image data to the autoencoder; and
determining whether the traffic data for each session is abnormal by comparing the BCE with the preset threshold value.
5. The method of claim 4 , wherein the determining whether the traffic data for each session is abnormal by comparing the BCE with the preset threshold value includes determining that the traffic data for each session is abnormal when the BCE is equal to or greater than the preset threshold value.
6. The method of claim 4 , wherein the determining whether the traffic data for each session is abnormal by comparing the BCE with the preset threshold value further includes determining that the traffic data for each session is not abnormal when the BCE is less than the preset threshold value.
7. The method of claim 2 , wherein the generating data clusters according to each of the plurality of different encoders using a clustering algorithm based on the traffic data for learning and the plurality of different encoders when the traffic data for each session is determined abnormal includes:
inputting the divided traffic data for learning to the plurality of different encoders and storing each of output data of the plurality of different encoders; and
generating the data clusters according to each of the plurality of different encoders by applying the clustering algorithm to each of the stored output data of the plurality of different encoders.
8. The method of claim 1 , wherein the re-determining whether the traffic data for each session is abnormal based on whether the output data included in the data cluster according to each of the different encoders includes re-determining that the traffic data for each session is abnormal when the output data is not included in any one of the data clusters according to the plurality of different encoders.
9. An apparatus for detecting anomaly traffic, comprising:
a processor; and
a memory configured to store at least one command executed through the processor,
wherein the at least one command including
generating a plurality of different encoders on the basis of traffic data for learning,
generating a plurality of pieces of image data on the basis of the traffic data for each session, which is a detection target,
determining whether the traffic data for each session is abnormal based on binary cross entropy (BCE) of the plurality of pieces of image data and a preset threshold value,
generating data clusters according to each of the plurality of different encoders using a clustering algorithm based on the traffic data for learning and the plurality of different encoders when the traffic data for each session is determined normal based on the BCE,
generating output data by inputting the plurality of image data to each of the plurality of different encoders for each image data sequentially, and
re-determining whether the traffic data for each session is abnormal based on whether the output data included in the data cluster according to each of the different encoders,
wherein the plurality of different encoders outputs different data for one image data, and
wherein the generating a plurality of pieces of image data comprises:
converting each character constituting a character string included in the traffic data for each session into a plurality of one-hot vectors in a reverse order;
generating a matrix by combining the plurality of one-hot vectors; and
generating an image representing the location of each character in the matrix.
10. The apparatus of claim 9 , wherein the at least one command includes
dividing the traffic data for learning into a preset unit,
converting the divided traffic data for learning into a plurality of pieces of image data for learning, and
generating the plurality of different encoders on the basis of the plurality of pieces of image data for learning.
11. The apparatus of claim 10 , wherein the at least one command includes
learning the plurality of pieces of image data for learning through an autoencoder based on an inception-residual network-v2 using convolutional neural networks to generate the plurality of different encoders.
12. The apparatus of claim 11 , wherein the at least one command includes
calculating BCE by inputting the plurality of pieces of image data to the autoencoder, and
determining whether the traffic data for each session is abnormal by comparing the BCE with a preset threshold value.
13. The apparatus of claim 12 , wherein the at least one command includes determining that the traffic data for each session is abnormal when the BCE is equal to or greater than the preset threshold value.
14. The apparatus of claim 12 , wherein the at least one command includes determining that the traffic data for each session is not abnormal when the BCE is less than the preset threshold value.
15. The apparatus of claim 10 , wherein the at least one command includes
inputting the divided traffic data for learning to the plurality of different encoders and store each of output data of the plurality of different encoders, and
generating the data clusters according to the plurality of different encoders by applying the clustering algorithm to each of the stored output data of the plurality of different encoders.
16. The apparatus of claim 9 , wherein the at least one command includes re-determining that the traffic data for each session is abnormal when the output data is not included in any one of the data clusters according to the plurality of different encoders.