Methods and apparatus to generate anomaly detection datasets

Abstract

Example methods and apparatus to generate anomaly detection datasets are disclosed. An example method to generate an anomaly detection dataset for training a machine learning model to detect real world anomalies includes receiving a user definition of an anomaly generator function, executing, with a processor, the anomaly generator function to generate user-defined anomaly data, and combining the user-defined anomaly data with nominal data to generate the anomaly detection dataset.

What is claimed is: 
     
         1 . A non-transitory computer-readable storage medium comprising instructions that, when executed, cause a machine to at least:
 obtain a user definition of a function to generate a normal data series;   determine a timeline of data based on the function;   determine a time scale associated with anomaly generation;   add anomaly data to the timeline of data based on the time scale;   cause storage of the timeline as an anomaly detection dataset with the anomaly data in a data map including timestamps associated with the anomaly data; and   train a machine learning classifier utilizing the anomaly detection dataset.   
     
     
         2 . The non-transitory computer-readable storage medium of  claim 1 , wherein the machine learning classifier is a deep-learned-based classifier. 
     
     
         3 . The non-transitory computer-readable storage medium of  claim 1 , wherein the instructions, when executed, cause the machine to generate random anomaly data. 
     
     
         4 . The non-transitory computer-readable storage medium of  claim 1 , wherein adding the anomaly data is based on an anomaly probability parameter. 
     
     
         5 . The non-transitory computer-readable storage medium of  claim 1 , wherein the instructions, when executed, cause the machine to generate anomaly data based on a user-defined function. 
     
     
         6 . The non-transitory computer-readable storage medium of  claim 1 , wherein the anomaly data and the normal data include respective data slices, wherein the instructions, when executed, cause the machine to splice the data slices together to combine the anomaly data with the normal data. 
     
     
         7 . The non-transitory computer-readable storage medium of  claim 1 , wherein the user definition of the function includes an executable file. 
     
     
         8 . An apparatus comprising:
 memory;   computer executable instructions;   programmable circuitry to execute the computer executable instructions to:
 obtain a user definition of a function to generate a normal data series; 
 determine a timeline of data based on the function; 
 determine a time scale associated with anomaly generation; 
 add anomaly data to the timeline of data based on the time scale; 
 cause storage of the timeline as an anomaly detection dataset with the anomaly data in a data map including timestamps associated with the anomaly data; and 
 train a machine learning classifier utilizing the anomaly detection dataset. 
   
     
     
         9 . The apparatus of  claim 8 , wherein the machine learning classifier is a deep-learned-based classifier. 
     
     
         10 . The apparatus of  claim 8 , programmable circuitry is to execute the computer executable instructions to generate random anomaly data. 
     
     
         11 . The apparatus of  claim 8 , wherein adding the anomaly data is based on an anomaly probability parameter. 
     
     
         12 . The apparatus of  claim 8 , programmable circuitry is to execute the computer executable instructions to generate anomaly data based on a user-defined function. 
     
     
         13 . The apparatus of  claim 8 , wherein the anomaly data and the normal data include respective data slices, programmable circuitry is to execute the computer executable instructions to splice the data slices together to combine the anomaly data with the normal data. 
     
     
         14 . The apparatus of  claim 8 , wherein the user definition of the function includes an executable file. 
     
     
         15 . A method comprising:
 obtaining a user definition of a function to generate a normal data series;   determining a timeline of data based on the function;   determining a time scale associated with anomaly generation;   adding anomaly data to the timeline of data based on the time scale;   causing storage of the timeline as an anomaly detection dataset with the anomaly data in a data map including timestamps associated with the anomaly data; and   training a machine learning classifier utilizing the anomaly detection dataset.   
     
     
         16 . The method of  claim 15 , wherein the machine learning classifier is a deep-learned-based classifier. 
     
     
         17 . The method of  claim 15 , further comprising generating random anomaly data. 
     
     
         18 . The method of  claim 15 , wherein adding the anomaly data is based on an anomaly probability parameter. 
     
     
         19 . The method of  claim 15 , further comprising generating anomaly data based on a user-defined function. 
     
     
         20 . The method of  claim 15 , wherein the anomaly data and the normal data include respective data slices, further comprising splicing the data slices together to combine the anomaly data with the normal data. 
     
     
         21 . The method of  claim 15 , wherein the user definition of the function includes an executable file.