# US 2021/0342204 A1 · Computer-based systems involving an engine and tools for incident prediction using machine learning and methods of use thereof

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 2021/0342204 A1
- Document ID: US20210342204A1
- Status: Published application
- Kind: A1
- Application number: 16/862,413
- Filed: 2020-04-29
- Published: 2021-11-04
- Earliest claimed: 2020-04-29
- Assignees: Capital One Services, LLC
- Applicants: Capital One Services, LLC
- Inventors: Suvro Choudhury; Cameron Stinson; Kellie Meshell; Henry Real
- Web page: https://patentgrep.com/patent/US20210342204A1

## CPC

- G06F 11/004: {Error avoidance (G06F11/07 and subgroups take precedence)}
- G06F 11/0778: {Dumping, i.e. gathering error/state information after a fault for later diagnosis}
- G06F 11/008: {Reliability or availability analysis}
- G06F 11/0709: {in a distributed system consisting of a plurality of standalone computer nodes, e.g. clusters, client-server systems}
- G06F 11/0793: {Remedial or corrective actions (recovery from an exception in an instruction pipeline G06F9/3861; by retry G06F11/1402; for recovering from a failure of a protocol instance or entity H04L69/40)}
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)

## Abstract

Systems and methods associated with incident prediction using machine learning techniques are disclosed. In one embodiment, an exemplary method may comprise obtaining current raw log data from at least one application log of at least one software application, converting the current raw log data into current tabular log data, applying one or more sampling techniques to the current tabular log data to form current balanced log data, the current balanced log data including incidents of failures, applying one or more machine learning techniques to the current balanced log data to generate an application failure predictive model, and predicting, based on future balanced log data, at least one future failure of the software application using the application failure predictive model.

## Claims

### Claim 1

(canceled)

### Claim 2 (depends on 14)

The method of claim 14, wherein the one or more sampling techniques comprise an undersampling technique.

### Claim 3 (depends on 2)

The method of claim 2, wherein the undersampling technique comprises:
parsing the current tabular log data into majority class data and minority class data;
reducing the majority class data to a size commensurate with the size of the minority class data; and
generating the current balanced log data using the reduced-size majority class data and the minority class data.

### Claim 4 (depends on 14)

The method of claim 14, wherein applying the one or more machine learning techniques to the current balanced log data comprises utilizing logistic regression to determine relationships between at least one dependent variable of the current balanced log data and one or more independent variables of the current balanced log data.

### Claim 5 (depends on 4)

The method of claim 4 wherein the logistic regression comprises a least absolute shrinkage and selection operator (LASSO) logistic regression that performs variable selection and regularization of the current balanced log data.

### Claim 6 (depends on 14)

The method of claim 14 further comprising:
splitting the current balanced log data into training data and testing data, wherein the training data and the testing data are utilized to verify the model's integrity.

### Claim 7 (depends on 14)

The method of claim 14, wherein the applying the one or more sampling techniques to the current tabular log data comprises one hot encoding the current tabular log data.

### Claim 8 (depends on 14)

The method of claim 14, further comprising:
configuring the tabular data into a readable tabular format, wherein one or both of the applying the one or more sampling techniques and/or the applying the one or more machine learning techniques are performed on the readable tabular format of the tabular data.

### Claim 9

A computer-implemented method comprising:
obtaining, by at least one processor, current raw log data from at least one application log of at least one software application;
converting, by the at least one processor, the current raw log data into current tabular log data;
configuring the current tabular log data into a readable tabular format;
applying, by the at least one processor, one or more sampling techniques to the current tabular log data to form current balanced log data, wherein the current balanced log data includes previous incidents of failures of the at least one software application;
applying, by the at least one processor, one or more machine learning techniques to the current balanced log data to generate an application failure predictive model based at least in part on the previous incidents of failures of the at least one software application;
wherein one or both of the applying the one or more sampling techniques and/or the applying the one or more machine learning techniques are performed on the readable tabular format of the current tabular log data;
generating, by the at least one processor, at least one indication of at least one future failure of the software application based on future balanced log data and the application failure predictive model; and
causing, by the at least one processor, based on the at least one indication, at least one remedial activity to be performed to or with the software application to mitigate or avoid the at least one future failure of the software application;
wherein the configuring the tabular data into the readable tabular format comprises generating the dictionary and fields that provide additional details regarding the log data, wherein the dictionary and fields are stored with the readable tabular format of the tabular data in a virtual environment.

### Claim 10 (depends on 9)

The method of claim 9, further comprising:
generating a dictionary for the current tabular log data, the dictionary comprising a dictionary list and dictionary entries for each element of log data.

### Claim 11

A computer-implemented method comprising:
obtaining, by at least one processor, current raw log data from at least one application log of at least one software application;
converting, by the at least one processor, the current raw log data into current tabular log data;
generating a dictionary for the current tabular log data, the dictionary comprising a dictionary list and dictionary entries for each element of log data;
configuring the current tabular data into the readable tabular format, via:
returning a dictionary list;
scanning a row of the dictionary list to generate keys that are used as column headers in the readable tabular format;
writing, to a comma-separated value (CSV) file, the keys that correspond to the column headers; and
writing rows from the dictionary list to fill the CSV file;
applying, by the at least one processor, one or more sampling techniques to the current tabular log data to form current balanced log data, wherein the current balanced log data includes previous incidents of failures of the at least one software application;
applying, by the at least one processor, one or more machine learning techniques to the current balanced log data to generate an application failure predictive model based at least in part on the previous incidents of failures of the at least one software application;
wherein one or both of the applying the one or more sampling techniques and/or the applying the one or more machine learning techniques are performed on the readable tabular format of the current tabular log data;
generating, by the at least one processor, at least one indication of at least one future failure of the software application based on future balanced log data and the application failure predictive model; and
causing, by the at least one processor, based on the at least one indication, at least one remedial activity to be performed to or with the software application to mitigate or avoid the at least one future failure of the software application.

### Claim 12 (depends on 14)

The method of claim 14, further comprising:
transforming the raw log data into a readable format; and
generating input date, time, logType, and message fields for each instance of the transformed raw log data.

### Claim 13 (depends on 17)

The method of claim 17, further comprising:
generating the dictionary for the current tabular log data, the dictionary having a dictionary entry for each instance of log data, wherein dictionary entries are comprised of 5 or more of log name, date, time, timestamp, message, category, day of the week, month of the year, success, hour of the day, minute of the hour, and/or day of the month.

### Claim 14

A computer-implemented method comprising:
obtaining, by at least one processor, current raw log data from at least one application log of at least one software application;
converting, by the at least one processor, the current raw log data into current tabular log data;
creating a dictionary for the current tabular log data, the dictionary having a dictionary entry for each instance of log data; and
generating data in the dictionary via looping through every line of the log data and appending information regarding each said dictionary entry into the dictionary;
applying, by the at least one processor, one or more sampling techniques to the current tabular log data to form current balanced log data, wherein the current balanced log data includes previous incidents of failures of the at least one software application;
applying, by the at least one processor, one or more machine learning techniques to the current balanced log data to generate an application failure predictive model based at least in part on the previous incidents of failures of the at least one software application;
generating, by the at least one processor, at least one indication of at least one future failure of the software application based on future balanced log data and the application failure predictive model; and
causing, by the at least one processor, based on the at least one indication, at least one remedial activity to be performed to or with the software application to mitigate or avoid the at least one future failure of the software application.

### Claim 15 (depends on 14)

The method of claim 14, wherein the raw log data comprises outbound logs, inbound logs, and batch logs.

### Claim 16 (depends on 14)

The method of claim 14, further comprising:
creating a dictionary for the current tabular log data, the dictionary having a dictionary entry for each instance of log data;
updating one or more string names associated with one or more dictionary entries to numerical values, including:
processing a list of dictionary entries and a category variable;
looping-through a list of job names, from the one or more string names, to parse out a Business name, a Log Function name, a Business Function name, a Business Process name, and a numerical Log name;
looping-through a list of Business names, correlating each said Business name to a number, and returning a first numerical value;
looping-through a list of Log Function names, correlating each said Log Function name to a number, and returning a second numerical value;
looping-through a list of Business Function names, correlating each said Business Function name to a number, and returning a third numerical value;
looping-through a list of Business Process names, correlating each said Business Process name to a number, and returning a fourth numerical value; and
creating the numerical Log name via combining two or more of the first numerical value, the second numerical value, the third numerical value, and the fourth numerical value together to form the numerical Log name, wherein each said numerical Log name comprises a complete unique numerical identifier for each string name.

### Claim 17 (depends on 1)

A computer-implemented method of claim 1, further comprising:
obtaining, by at least one processor, current raw log data from at least one application log of at least one software application;
converting, by the at least one processor, the current raw log data into current tabular log data:
updating a dictionary associated with the current tabular log data, including:
processing dictionary list entries comprised of an argument field, an index field, a Dictionary List field, a Log Category field, a Business Name field, a Log Function field, a Business Function field, a Business Process field, and a Log Name field; and
updating and adding fields comprising the Log Category field, the Business Name field, the Log Function field, the Business Function field, the Business Process field, and a Log Name to the dictionary, wherein the Log Name is a unique numerical identifier created by combining a plurality of fields into one unique value;
applying, by the at least one processor, one or more sampling techniques to the current tabular log data to form current balanced log data, wherein the current balanced log data includes previous incidents of failures of the at least one software application;
applying, by the at least one processor, one or more machine learning techniques to the current balanced log data to generate an application failure predictive model based at least in part on the previous incidents of failures of the at least one software application;
generating, by the at least one processor, at least one indication of at least one future failure of the software application based on future balanced log data and the application failure predictive model; and
causing, by the at least one processor, based on the at least one indication, at least one remedial activity to be performed to or with the software application to mitigate or avoid the at least one future failure of the software application.

### Claim 18

(canceled)

### Claim 19

(canceled)

### Claim 20

(canceled)

### Claim 21 (depends on 17)

The method of claim 17, further comprising:
generating a dictionary for the current tabular log data, the dictionary comprising a dictionary list and dictionary entries for each element of log data.

### Claim 22 (depends on 17)

The method of claim 17, wherein the one or more sampling techniques comprise an undersampling technique.

### Claim 23 (depends on 9)

The method of claim 9, wherein applying the one or more machine learning techniques to the current balanced log data comprises:
utilizing logistic regression to determine relationships between at least one dependent variable of the current balanced log data and one or more independent variables of the current balanced log data.

### Claim 24 (depends on 11)

The method of claim 11, further comprising:
splitting the current balanced log data into training data and testing data, wherein the training data and the testing data are utilized to verify the model's integrity.
