METHOD AND SYSTEM FOR ADAPTIVELY REMOVING OUTLIERS FROM DATA USED IN TRAINING OF PREDICTIVE MODELS

• US 20180081913A1
• Filed: 09/18/2017
• Published: 03/22/2018
• Est. Priority Date: 09/16/2016
• Status: Active Grant

• Current Assignee: Oracle International Corporation (Oracle Corporation)
• Original Assignee: Oracle International Corporation (Oracle Corporation)
• Inventors: SULEIMAN, Yaser I., ZOLL, Michael, BASU, Subhransu, PRUSCINO, Angelo, LOHWASSER, Wolfgang, MIYOSHI, Wataru, BREIDT, Thomas, HERTER, Thomas, THIELEN, Klaus, KUMAR, Sahil
• Granted Patent: US 11,308,049 B2
• Time in Patent Office: Days
• CPC Class Codes: G06F 11/3409 for performance assessment
• CPC Class Codes: G06F 11/3452 Performance evaluation by statistical analysis
• CPC Class Codes: G06F 16/215 Improving data quality; Data cleansing, e.g. de-duplication, removing invalid entries or correcting typographical errors
• CPC Class Codes: G06F 18/214 Generating training patterns; Bootstrap methods, e.g. bagging or boosting
• CPC Class Codes: G06N 20/00 Machine learning
• CPC Class Codes: G06N 20/20 Ensemble learning
• CPC Class Codes: G06N 7/01 Probabilistic graphical models, e.g. probabilistic networks
• CPC Class Codes: H04L 41/069 using logs of notifications; Post-processing of notifications
• CPC Class Codes: H04L 41/0695 the faulty arrangement being the maintenance, administration or management system
• CPC Class Codes: H04L 41/0823 characterised by the purposes of a change of settings, e.g. optimising configuration for enhancing reliability for optimising operational conditions of wireless networks H04W24/02
• CPC Class Codes: H04L 41/142 using statistical or mathematical methods
• CPC Class Codes: H04L 43/02 Capturing of monitoring data

Abstract

Described is an improved approach to remove data outliers by filtering out data correlated to detrimental events within a system. One or more detrimental even conditions are defined to identify and handle abnormal transient states from collected data for a monitored system.

Claims

1. A method for generating training data for a machine learning system, comprising:

collecting data from a monitored system;

identifying one or more filter conditions, where the one or more filter conditions correspond to identified events that occur within the monitored system;

filtering the data pursuant to the one or more filter conditions, wherein the data is filtered by scanning candidate datasets to identify a datapoint temporally associated with the identified events, and the datapoint is not placed within a filtered dataset; and

performing model training with the filtered data.

2. The method of claim 1, further comprising removing at least one of a preceding datapoint or a subsequent datapoint from the data to generate the filtered dataset.

3. The method of claim 1, wherein a scope is defined for the one or more filter conditions, and the scope identifies an extent to which filtering is applied upon detection of the identified events.

4. The method of claim 3, wherein the scope corresponds to at least one of a global scope where filtering is applied to all levels of the monitored system'"'"'s component hierarchy, a local scope where the scope of filtering is applied only to a local level of a node or instance, or an intermediate scope where filtering is applied to a common level of a component hierarchy of the monitored system.

5. The method of claim 1, wherein the filter condition is defined with a declaration that includes a signal name, a scope, a formula, or an apply period.

6. The method of claim 1, wherein filtering is applied as a pre-processing action when collecting the data, as a post-processing action after collection of the data, or a combination thereof.

7. The method of claim 1, wherein a predictive model is generated from the model training with the training data, the predictive model being applied to perform health monitoring of a database cluster.

8. A system for generating training data for a machine learning system, comprising:

a processor;

a memory for holding programmable code; and

wherein the programmable code includes instructions for collecting data from a monitored system;

identifying one or more filter conditions, where the one or more filter conditions correspond to identified events that occur within the monitored system;

filtering the data pursuant to the one or more filter conditions, wherein the data is filtered by scanning candidate datasets to identify a datapoint that is a manifestation of the identified events any of the given detrimental events, and the datapoint is not placed within a filtered dataset; and

performing model training with the filtered data.

9. The method of claim 8, wherein the programmable code further includes instructions for removing at least one of a preceding datapoint or an antecedent datapoint from the data to generate the filtered dataset.

10. The method of claim 8, wherein a scope of defined for the one or more filter conditions, and the scope identifies an extent to which filtering is applied upon detestation of the identified events.

11. The method of claim 10, wherein the scope corresponds to at least one of a global scope where filtering is applied to all levels of the monitored system'"'"'s component hierarchy, a local scope where the scope of filtering is applied only to a local level of a node or instance, or an intermediate scope where filtering is applied to a common level of the monitored system'"'"'s component hierarchy.

12. The method of claim 8, wherein the filter condition is defined with a declaration that includes a signal name, a scope, a formula, or an apply period.

13. The method of claim 8, wherein filtering is applied as a pre-processing action when collecting the data, as a post-processing action after collection of the data, or a combination thereof.

14. The method of claim 8, wherein a predictive model is generated from the model training with the training data, the predictive model being applied to perform health monitoring of a database cluster.

15. A computer program product embodied on a computer readable medium, the computer readable medium having stored thereon a sequence of instructions which, when executed by a processor, executes a method comprising:

collecting data from a monitored system;

identifying one or more filter conditions, where the one or more filter conditions correspond to identified events that occur within the monitored system;

filtering the data pursuant to the one or more filter conditions, wherein the data is filtered by scanning candidate datasets to identify a datapoint that is a manifestation of the identified events any of the given detrimental events, and the datapoint is not placed within a filtered dataset; and

performing model training with the filtered data.

16. The computer program product of claim 15, wherein the sequence of instructions, when executed by a processor, further executes the step of removing at least one of a preceding datapoint or an antecedent datapoint from the data to generate the filtered dataset.

17. The computer program product of claim 15, wherein a scope of defined for the one or more filter conditions, and the scope identifies an extent to which filtering is applied upon detestation of the identified events.

18. The computer program product of claim 17, wherein the scope corresponds to at least one of a global scope where filtering is applied to all levels of the monitored system'"'"'s component hierarchy, a local scope where the scope of filtering is applied only to a local level of a node or instance, or an intermediate scope where filtering is applied to a common level of the monitored system'"'"'s component hierarchy.

19. The computer program product of claim 15, wherein the filter condition is defined with a declaration that includes a signal name, a scope, a formula, or an apply period.

20. The computer program product of claim 15, wherein filtering is applied as a pre-processing action when collecting the data, as a post-processing action after collection of the data, or a combination thereof.

21. The computer program product of claim 15, wherein a predictive model is generated from the model training with the training data, the predictive model being applied to perform health monitoring of a database cluster.