A System And A Method For Detecting Anomalous Patterns In A Network

Cognizant Technology Solutions India Pvt. Ltd.
Granted App No 201941043497 Patent 520235 Communication

Patent information

IP India · live record
Application number
201941043497
Filing date
25 October 2019
Status
Granted
Legal status
Inforce
Publication number
18/2021
Publication type
INA
Invention field
Communication
Patent number
520235
Grant date
05 March 2024
Next renewal due
25 October 2034

Abstract

The present invention discloses a system and a method for detecting anomalous patterns in a network such as a LAN, WAN, MAN, internet of things(Iot), cloud networks, or any other network. In operation, the system and method of the present invention determines a generic pattern of behavior associated with a plurality of anomaly classes based on a plurality of feature values using reinforcement learning technique. The generic pattern is fixed as a boundary for each of the plurality of anomaly classes and is representative of behavior which substantially simulates the network behavior on attack by any of the plurality of anomaly classes. Further, the present invention, provides for updating the generic pattern using reinforcement learning. The updated generic pattern is implemented to analyze and detect anomalous behavior in the incoming network traffic in real time.

Applicants

1
India
Cognizant Technology Solutions India Pvt. Ltd. Techno Complex, No. 5/535, Old Mahabalipuram Road, Okkiyam Thoraipakkam, Chennai 600 097, Tamil Nadu, India

Inventors

5
India
Lakshmanan BabuG1, Udayam Flats, No:5 Teachers Colony West Canal Bank Road, Kotturpuram, Chennai – 600085, Tamil Nadu, India
India
S. Vinoth3/218, V.S. Nagar First Street, Pasupathikoil, Thanjavur (Dt.) – 614206, Tamil Nadu, India
India
V. SrihariNo-44, Police Line First Street, Tirupur – 641601 (Near Indian Bank), Tamil Nadu, India
India
C. RohithCherikkallil House, Vattamkulam, Malappuram (Dt.) - 679578, Kerala, India
India
D. KeerthikaNo:52, Sree Laksshmi Nagar, Vanagaram, Near Porur Toll Gate, Chennai – 600095, Tamil Nadu, India

Specification

Show full textCollapse

1. A method for detecting anomalous behavior patterns in a network, wherein the method is implemented by at least one processor (114) executing program instructions stored in a memory(116), the method comprising: extracting, by the processor(114), a plurality of feature values associated with predetermined features from a first collection of one or more datasets associated with a plurality of anomaly classes, wherein the predetermined features are individual measurable characteristics of network behavior and user behavior in the presence or absence of an anomaly; determining, by the processor(114), a generic pattern of behavior associated with the plurality of anomaly classes based on the extracted feature values, wherein the generic pattern is representative of behavior which substantially simulates feature values on attack by any of the plurality of anomaly classes; updating, by the processor(114), the determined generic pattern based on the analysis of performance of the determined generic pattern based on a second collection of one or more datasets associated with the plurality of anomaly classes and normal behavior classes; and detecting, by the processor(114), anomalous behavior pattern in a real-time traffic based on the updated generic pattern. 2. The method as claimed in claim 1, wherein, a notification indicating abnormal traffic is generated if the feature values associated with the real-time traffic is abnormal. 3. The method as claimed in claim 1, wherein the updated generic pattern is further updated based on cumulative rewards generated based on each correct notification. 4. The method as claimed in claim 1, wherein extracting the plurality of feature values from the first collection of one or

more datasets comprises refining the first collection of one or more datasets using a first set of rules and using at least one of: a probability distribution technique and traffic pattern recognition techniques on the refined first collection of one or more datasets to extract the plurality of feature values, wherein the first set of rules comprises: categorizing a plurality of files included in the first collection of one or more datasets into network traffic files and user behavior files, wherein the network traffic files include data associated with network anomaly class and normal traffic class, and the user behavior files include data associated with user behavior anomalies and normal user behavior; segmenting each of the categorized datasets, wherein the network traffic files and the user behavior files are segmented into small chunks network packets and user logs respectively based on time of collection of data within the files using Long-Short-Term memory (LSTM) technique; removing improper datasets from the segmented datasets, wherein each of the network traffic file and each of the user behavior file is processed to remove improper network packets and improper user logs respectively; and arranging the network packets and the user logs in a chronological order. 5. The method as claimed in claim 1, wherein the plurality of feature values associated with the predetermined features are extracted from refined network packets, using a probability distribution technique, the probability distribution technique selected from a group of Levy, Cauchy, Pareto, Weibull and Lognormal. 6. The method as claimed in claim 1, wherein determining the generic pattern of behavior comprises: labeling the extracted plurality of feature values for each of a plurality of network traffic files and user behavior

files included in the first collection of one or more datasets as abnormal or normal using supervised learning; feeding each file with labeled features as an input to a deep Q learning technique; identifying the fed files as normal or abnormal based on the feature values by performing a random identification action on the fed files using the deep Q learning technique, wherein the random identification action is evaluated based on the labeling, wherein a positive reward is provided if the random action is same as the label and a negative reward is provided if the random action is not same as the label, and computing a total cumulative reward based on the total number of correct actions for the plurality of files of respective anomaly classes in the first collection of one or more datasets; and evaluating individual patterns for respective anomaly classes based on identification of fed files, and determining the generic pattern of behavior associated with the plurality of anomaly classes from the individual pattern of anomaly classes and feature values associated with respective anomaly classes, the generic pattern is fixed as a boundary for each of the plurality of anomaly classes. 7. The method as claimed in claim 1, wherein updating and analysis of the determined generic pattern is performed using a second set of rules, wherein the second set of rules comprises: refining the second collection of one or more datasets, wherein a plurality of files included in the second collection of one or more datasets are categorized into network traffic files and user behavior files, segmented into network packets and user logs, and processed to remove improper network packets and user logs, wherein the processed network packets and user logs are arranged in a chronological order respectively;

extracting a plurality of feature values associated with the predetermined features from the plurality of files included in the second collection of one or more datasets; identifying the feature values associated with each file included in the second collection of one or more datasets as normal or abnormal by analyzing the plurality of extracted feature values based on the determined generic pattern using reinforcement learning, wherein each file included in the second collection of one or more datasets is fed as an input to a deep Q learning technique; evaluating each identification action based on the determined generic pattern and generating rewards for each correct identification action; and updating the generic pattern based on cumulative rewards generated based on the total number of correct actions for the plurality of files of respective anomaly classes in the second collection of one or more datasets. 8. The method as claimed in claim 1, wherein anomalous behavior pattern in the network is detected by analyzing the real-time traffic using a third set of rules, wherein the third set of rules comprises: refining the incoming real-time traffic, wherein a plurality of files included in the real-time traffic are categorized into network traffic files and user behavior files, segmented into network packets and user logs, and processed to remove improper network packets and user logs, wherein the processed network packets and user logs are arranged in a chronological order respectively; extracting a plurality feature values associated with the predetermined features from each of the plurality of files included in the incoming real-time traffic; identifying the respective feature values of the refined files as normal or abnormal by analyzing the extracted plurality

of feature values based on the updated generic pattern using reinforcement learning, wherein each of refined files included in the incoming real-time traffic is fed as an input to a deep Q learning technique; and evaluating each identification action based on the updated generic pattern and generating rewards for each correct identification action. 9. A system (108) for detecting anomalous behavior patterns in a network, the system (108) comprising: a memory (116) storing program instructions; a processor (114) configured to execute program instructions stored in the memory (116); and an anomaly detection engine (112) in communication with the processor and configured to: extract a plurality of feature values associated with predetermined features from a first collection of one or more datasets associated with a plurality of anomaly classes, wherein the predetermined features are individual measurable characteristics of network behavior and user behavior in the presence or absence of an anomaly; determine a generic pattern of behavior associated with the plurality of anomaly classes based on the extracted feature values, wherein the generic pattern is representative of behavior which substantially simulates feature values on attack by any of the plurality of anomaly classes; update the determined generic pattern based on the analysis of performance of the determined generic pattern based on a second collection of one or more datasets associated with the plurality of anomaly classes and normal behavior classes; and detect anomalous behavior pattern in a real-time traffic based on the updated generic pattern.

10. The system (108) as claimed in claim 9, wherein, a notification indicating abnormal traffic is generated if the feature values associated with the real-time traffic is abnormal. 11. The system (108) as claimed in claim 9, wherein the updated generic pattern is further updated based on cumulative rewards generated based on each correct notification. 12. The system (108) as claimed in claim 9, wherein extracting the plurality of feature values from the first collection of one or more datasets comprises refining the first collection of one or more datasets using a first set of rules and using at least one of: a probability distribution technique and traffic pattern recognition techniques on the refined first collection of one or more datasets to extract the plurality of feature values. 13. The system (108) as claimed in claim 9, wherein the anomaly detection engine (112) comprises a data segmentation unit (120) in communication with the processor (114), said data segmentation unit (120) configured to refine the first collection of datasets, second collection of one or more datasets and the incoming real-time traffic using a first set of rules, wherein the first set of rules comprises a. categorizing a plurality of files included in the first collection of one or more datasets into network traffic files and user behavior files, wherein the network traffic files include data associated with network anomaly class and normal traffic class, and the user behavior files include data associated with user behavior anomalies and normal user behavior; b. segmenting each of the categorized datasets, wherein the network traffic files and the user behavior files are segmented into small chunks network packets and user logs respectively based on time of collection of data within the files using Long- Short-Term memory (LSTM) technique;

c. removing improper datasets from the segmented datasets, wherein each of the network traffic file and each of the user behavior file is processed to remove improper network packets and improper user logs respectively; d. arranging the network packets and the user logs in a chronological order; and e. repeating steps a-d for the second collection of one or more datasets and the incoming real-time traffic. 14. The system (108) as claimed in claim 9, wherein the anomaly detection engine (112) comprises a feature extraction unit (122) in communication with the processor (114), said feature extraction unit (122) configured to extract the plurality of feature values associated with network anomaly classes from refined network packets using a probability distribution technique and a traffic pattern recognition techniques, wherein the network traffic files associated with the refined network packets are fitted into one or more probability distribution techniques, and the parameters of the distribution are used as distribution features. 15. The system (108) as claimed in claim 9, wherein the anomaly detection engine comprises an identification unit in communication with the processor, said identification unit configured to determine the generic pattern of behavior by: labeling the extracted plurality of feature values for each of a plurality of network traffic files and user behavior files included in the first collection of one or more datasets as abnormal or normal using supervised learning; feeding each file with labeled features as an input to a deep Q learning technique; identifying the fed files as normal or abnormal based on the feature values by performing a random identification action on the fed files using the deep Q learning technique, wherein the random identification action is evaluated based on the labeling,

wherein a positive reward is provided if the random action is same as the label and a negative reward is provided if the random action is not same as the label, and computing a total cumulative reward based on the total number of correct actions for the plurality of files of respective anomaly classes in the first collection of one or more datasets; and evaluating individual patterns for respective anomaly classes based on identification of fed files, and determining the generic pattern of behavior associated with the plurality of anomaly classes from the individual pattern of anomaly classes and feature values associated with respective anomaly classes, the generic pattern is fixed as a boundary for each of the plurality of anomaly classes. 16. The system (108) as claimed in claim 9, wherein the anomaly detection engine (112) comprises an identification unit (124) in communication with the processor, said identification unit (124) configured to analyze performance of the determined generic pattern and update said generic pattern using a second set of rules, wherein the second set of rules comprises: refining the second collection of one or more datasets, wherein a plurality of files included in the second collection of one or more datasets are categorized into network traffic files and user behavior files, segmented into network packets and user logs, and processed to remove improper network packets and user logs, wherein the processed network packets and user logs are arranged in a chronological order respectively; extracting a plurality of feature values associated with the predetermined features from the plurality of files included in the second collection of one or more datasets; identifying the feature values associated with each file included in the second collection of one or more datasets as normal or abnormal by analyzing the plurality of extracted features based on the determined generic pattern using reinforcement learning, wherein each file included in the second

collection of one or more datasets is fed as an input to a deep Q learning technique; evaluating each identification action based on the determined generic pattern and generating rewards for each correct identification action; and updating the generic pattern based on cumulative rewards generated based on the total number of correct actions for the plurality of files of respective anomaly classes in the second collection of one or more datasets. 17. The system (108) as claimed in claim 9, wherein the anomaly detection engine (112) comprises an identification unit (124) in communication with the processor, said identification unit (124) configured to detect anomalous behavior pattern in the network by analyzing real-time traffic using a third set of rules, wherein the third set of rules comprises: refining the incoming real-time traffic, wherein a plurality of files included in the real-time traffic are categorized into network traffic files and user behavior files, segmented into network packets and user logs, and processed to remove improper network packets and user logs, wherein the processed network packets and user logs are arranged in a chronological order respectively; extracting a plurality feature values associated with the predetermined features from each of the plurality of files included in the incoming real-time traffic; identifying the respective feature values of the refined files as normal or abnormal by analyzing the extracted plurality of feature values based on the updated generic pattern using reinforcement learning, wherein each of refined files included in the incoming real-time traffic is fed as an input to a deep Q learning technique; and

evaluating each identification action based on the updated generic pattern and generating rewards for each correct identification action.

Documents

Sign in to open
Application documents
1 201941043497-STATEMENT OF UNDERTAKING (FORM 3) [25-10-2019(online)].pdf 25 Oct 2019
2 201941043497-PROOF OF RIGHT [25-10-2019(online)].pdf 25 Oct 2019
3 201941043497-POWER OF AUTHORITY [25-10-2019(online)].pdf 25 Oct 2019
4 201941043497-FORM 1 [25-10-2019(online)].pdf 25 Oct 2019
5 201941043497-DRAWINGS [25-10-2019(online)].pdf 25 Oct 2019
6 201941043497-COMPLETE SPECIFICATION [25-10-2019(online)].pdf 25 Oct 2019
7 201941043497-FORM 18 [30-10-2019(online)].pdf 30 Oct 2019
8 Correspondence by Agent_Form1,Form26_04-11-2019.pdf 04 Nov 2019
9 201941043497-Request Letter-Correspondence [08-11-2019(online)].pdf 08 Nov 2019
10 201941043497-Form 1 (Submitted on date of filing) [08-11-2019(online)].pdf 08 Nov 2019
11 201941043497-FORM 3 [18-02-2020(online)].pdf 18 Feb 2020
12 201941043497-FER.pdf 17 Oct 2021
13 201941043497-PETITION UNDER RULE 137 [31-03-2022(online)].pdf 31 Mar 2022
14 201941043497-FORM 3 [31-03-2022(online)].pdf 31 Mar 2022
15 201941043497-FER_SER_REPLY [31-03-2022(online)].pdf 31 Mar 2022
16 201941043497-CLAIMS [31-03-2022(online)].pdf 31 Mar 2022
17 201941043497-ABSTRACT [31-03-2022(online)].pdf 31 Mar 2022
18 201941043497-PatentCertificate05-03-2024.pdf 05 Mar 2024
19 201941043497-IntimationOfGrant05-03-2024.pdf 05 Mar 2024
Search strategy
1searchstrategy201941043497E_05-08-2021.pdf

ERegister / renewals

Inforce
3rd Renewed 15 May 2024 Period 25 Oct 2021 – 25 Oct 2022 · CBR 32550 · Certificate 96305 ₹4,000
4th Renewed 15 May 2024 Period 25 Oct 2022 – 25 Oct 2023 · CBR 32550 · Certificate 96307 ₹4,000
5th Renewed 15 May 2024 Period 25 Oct 2023 – 25 Oct 2024 · CBR 32550 · Certificate 96310 ₹4,000
6th Renewed 15 May 2024 Period 25 Oct 2024 – 25 Oct 2025 · CBR 32550 · Certificate 96312 ₹4,000
7th Renewed 15 May 2024 Period 25 Oct 2025 – 25 Oct 2026 · CBR 32550 · Certificate 96313 ₹10,800
8th Renewed 15 May 2024 Period 25 Oct 2026 – 25 Oct 2027 · CBR 32550 · Certificate 96315 ₹10,800
9th Renewed 15 May 2024 Period 25 Oct 2027 – 25 Oct 2028 · CBR 32550 · Certificate 96317 ₹10,800
10th Renewed 15 May 2024 Period 25 Oct 2028 – 25 Oct 2029 · CBR 32550 · Certificate 96320 ₹10,800
11th Renewed 15 May 2024 Period 25 Oct 2029 – 25 Oct 2030 · CBR 32550 · Certificate 96322 ₹21,600
12th Renewed 15 May 2024 Period 25 Oct 2030 – 25 Oct 2031 · CBR 32550 · Certificate 96323 ₹21,600
13th Renewed 15 May 2024 Period 25 Oct 2031 – 25 Oct 2032 · CBR 32550 · Certificate 96326 ₹21,600
14th Renewed 15 May 2024 Period 25 Oct 2032 – 25 Oct 2033 · CBR 32550 · Certificate 96328 ₹21,600
15th Renewed 15 May 2024 Period 25 Oct 2033 – 25 Oct 2034 · CBR 32550 · Certificate 96329 ₹21,600