Granted Patent B2

US 12,493,719 · App. 18/610,945 · Granted Dec 9, 2025

Inventors: Shanmugaraja Krishnasamy Venugopal (Leander, TX); Steve VanWieren (Woodstock, GA)

Assignee: Dayforce US, Inc.

G06F21/6254

Abstract

Claims (348)

1 . A method to de-identify personally identifiable information (PII) within a structured dataset, the method comprising:

performing a PII identification operation by:

receiving the structured dataset, the structured dataset including a plurality of data elements;

performing a direct identifier detection operation by:

receiving a direct identifier threshold value;

based on the direct identifier threshold value, identifying direct identifiers within the structured dataset; and

creating a non-direct identifier structured dataset by removing the direct identifiers from the structured dataset;

performing a quasi-identifier detection operation by:

receiving a quasi-identifier threshold value;

based on the quasi-identifier threshold value, identifying quasi-identifiers within the non-direct identifier structured dataset; and

creating a non-quasi-identifier structured dataset by removing the quasi-identifiers from the non-direct identifier structured dataset;

performing a unique-value detection operation by:

receiving a unique value threshold value; and

based on the unique value threshold value, identifying unique values of data elements within the non-quasi identifier structured dataset;

performing a transformation operation by:

applying one or more transformation rules to transform each of the direct identifiers, the quasi-identifiers and the unique values into respective de-identified data elements; and

generating a de-identified structured dataset by replacing the direct identifiers, quasi-identifiers and the unique values within the structured dataset with the respective de-identified data elements.

2 . The method of claim 1 , further comprising:

performing a re-identification risk assessment operation by calculating a probability of re-identification score by using:

Probability

⁢

of

⁢

re

-

identification

⁢

score

=

1

Size

⁢

of

⁢

equivalence

⁢

class

wherein equivalence class is a number of unique groups of data that can be created within the de-identified structured dataset upon removal of the direct identifiers.

3 . The method of claim 2 , wherein performing the re-identification risk assessment operation further comprises:

receiving a re-identification threshold value;

comparing the probability of re-identification score to the re-identification threshold value;

upon determining that the probability of re-identification score meets or exceeds the re-identification threshold value:

generating an adjusted quasi-identifier threshold value by reducing the quasi-identifier threshold value by a predetermined amount; and

until determining that the probability of re-identification score is less than the re-identification threshold value, iteratively repeating performing the quasi-identifier detection operation, the unique-value detection operation, the transformation operation and the re-identification risk assessment operation after replacing the quasi-identifier threshold value with the adjusted quasi-identifier threshold value.

4 . The method of claim 1 , wherein identifying the direct identifiers in the structured dataset includes:

calculating a normalized entropy score for each data element of the plurality of data elements of the structured dataset;

for each data element of the plurality of data elements of the structured dataset:

comparing the normalized entropy score of the data element to the direct identifier threshold value; and

upon determining that the normalized entropy score meets or exceeds the direct identifier threshold value, identifying the data element as including direct identifiers.

5 . The method of claim 4 , wherein the normalized entropy score of a data element X is calculated using:

H

⁡

(

X

)

=

-

∑

i

=

1

N

p

⁡

(

x

i

)

*

log

2

⁢

p

⁡

(

x

i

)

max

⁢

entropy

⁢

score

wherein, H(X) is the normalized entropy score of the data element X, N is a total number of possible outcomes for the data element X, p(x i ) is a probability of each of the possible outcomes of the data element X occurring, Σ denotes a sum of all possible outcomes for the data element X and the max entropy score is calculated using the equation −Σ i=1 N p(x i )*log 2 p(x i ) where p(x i ) is equal to 1/N.

6 . The method of claim 1 , wherein identifying the quasi-identifiers in the non-direct identifier dataset includes:

calculating a normalized entropy score and an external influence score for each data element of the plurality of data elements of the non-direct identifier structured dataset;

creating a combined score each data element of the plurality of data elements of the non-direct identifier structured dataset by combining the normalized entropy score and the external influence score;

for each data element of the plurality of data elements of the non-direct identifier structured dataset:

comparing the combined score of the data element to the quasi-identifier threshold value; and

upon determining that the combined score meets or exceeds the quasi-identifier threshold value, identifying the data element as including quasi-identifiers.

7 . The method of claim 6 , wherein the normalized entropy score of a data element X is calculated using

H

⁡

(

X

)

=

-

∑

i

=

1

N

p

⁡

(

x

i

)

*

log

2

⁢

p

⁡

(

x

i

)

max

⁢

entropy

⁢

score

wherein, H(X) is the normalized entropy score of the data element X, N is a total number of possible outcomes for the data element X, p(x i ) is a probability of each of the possible outcomes of the data element X occurring, Σ denotes a sum of all possible outcomes for the data element X and the max entropy score is calculated using the equation −Σ i=1 N p(x i )*log 2 p(x i ) where p(x i ) is equal to 1/N; and

wherein the external influence score of a data element Y is calculated using:

External

⁢

influence

⁢

score

⁢

(

Y

)

=

1

-

N

E

(

T

-

Y

)

N

E

(

T

)

wherein, N E (T−Y) is a number of unique groups that can be formed after the data element Y is removed from the non-direct identifier structured dataset; and N E (T) is the maximum number of unique groups that can exist within the non-direct identifier structured dataset.

8 . The method of claim 1 , wherein identifying the unique values in the non-quasi-identifier dataset includes:

calculating an internal influence score for each value of a data element of the plurality of data elements of the non-quasi-identifier structured dataset:

comparing the internal influence score of the data element to unique value threshold value; and

upon determining that the internal influence score meets or exceeds the unique value threshold value, identifying the data element as including unique values.

9 . The method of claim 8 , wherein the internal influence score of a data element Z is calculated by performing a unique value count for each value in the data element Z.

10 . The method of claim 1 , wherein applying one or more transformation rules to transform each of the direct identifiers, the quasi-identifiers and the unique values includes:

selecting a transformation rule based on a data type and a category type of each of the direct identifiers, the quasi-identifiers and the unique values, wherein the data type includes: a numeric data type, a string data type, or a date data type; and

wherein the category type includes: a database identifier category or a non-database identifier category.

11 . A de-identification system comprising:

one or more processors; and

system memory, the system memory including instructions which, when executed by the processing unit, cause the one or more processors to:

perform a PII identification operation by:

receiving the structured dataset, the structured dataset including a plurality of data elements;

performing a direct identifier detection operation by:

receiving a direct identifier threshold value;

for each data element of the plurality of data elements of the structured dataset:

 calculating a first normalized entropy score;

 comparing the first normalized entropy score of the data column to the direct identifier threshold value; and

 upon determining that the first normalized entropy score meets or exceeds the direct identifier threshold value, identifying the data element of the structured dataset as including direct identifiers;

creating a non-direct identifier structured dataset by removing the direct identifiers from the structured dataset;

performing a quasi-identifier detection operation by:

receiving a quasi-identifier threshold value;

for each data element of the plurality of data elements of the non-direct identifier structured dataset:

calculating a second normalized entropy score;

calculating an external influence score;

creating a combined score by aggregating the second normalized entropy score to the external influence score;

comparing the combined score to the quasi-identifier threshold value; and

upon determining that the combined score meets or exceeds the quasi-identifier threshold value, identifying the data element of the non-direct identifier structured dataset as including quasi identifiers;

perform a transformation operation by:

applying one or more transformation rules to transform each of the direct identifiers, and the quasi-identifiers into respective de-identified data elements; and

generating a de-identified structured dataset by replacing the direct identifiers and the quasi-identifiers within the structured dataset with the respective de-identified data elements.

12 . The de-identification system of claim 11 , wherein the instructions further cause the one or more processors to:

perform a re-identification risk assessment operation by:

determining a probability of re-identification of personally identifiable information within the de-identified structured dataset;

upon determining that the probability of re-identification meets or exceeds a predetermined re-identification threshold value,

generating an adjusted quasi-identifier threshold value by reducing the quasi-identifier threshold value by a predetermined amount;

until determining that the probability of re-identification score is less than the re-identification threshold value, iteratively repeating the quasi-identifier detection operation, the transformation operation and the re-identification risk assessment operation after replacing the quasi-identifier threshold value with the adjusted quasi-identifier threshold value.

13 . The de-identification system of claim 11 , wherein the instructions further cause the one or more processors to:

create a non-quasi-identifier structured dataset by removing the quasi-identifiers from the non-direct identifier structured dataset;

receiving a unique value threshold value;

for each value of a data element of the plurality of data elements of the non-quasi-identifier structured dataset:

calculating an internal influence score;

comparing the internal influence score to the unique value threshold value; and

upon determining that the internal influence score meets or exceeds the unique value threshold value, identifying the data element of the non-quasi-identifier structured dataset as including unique values.

14 . The system of claim 11 , wherein the probability of re-identification score is calculated using:

Probability

⁢

of

⁢

re

-

identification

⁢

score

=

1

Size

⁢

of

⁢

equivalence

⁢

class

wherein equivalence class is a number of unique groups of data that can be created within the de-identified structured dataset upon removal of the direct identifiers.

15 . The system of claim 11 , wherein applying one or more transformation rules to transform each of the direct identifiers, and the quasi-identifiers into respective de-identified data elements includes:

selecting a transformation rule based on a data type and a category type of each of the direct identifiers and the quasi-identifiers,

wherein the data type includes: a numeric data type, a string data type, or a date data type; and

wherein the category type includes: a database identifier category or a non-database identifier category.

16 . The system of claim 15 , wherein applying the one or more transformation rules to transform each of the direct identifiers includes:

upon determining that the data type of the direct identifier is the numeric data type and the category type of the direct identifier is the non-database identifier category, applying a binning transformation rule or a generalization transformation rule;

upon determining that the data type of the direct identifier is the numeric data type and the category type of the direct identifier is the database identifier category, applying a numerical offset transformation rule;

upon determining that the data type of the direct identifier is the string data type, applying a format preserving encryption transformation rule;

upon determining that the data type of the direct identifier is the date data type and the category type of the direct identifier is the non-database identifier category, applying a binning transformation rule or a generalization transformation rule; and

upon determining that the data type of the direct identifier is the date data type and the category type of the direct identifier is the database identifier category, applying a date offset transformation rule.

17 . The system of claim 15 , wherein applying the one or more transformation rules to transform each of the quasi-identifiers includes:

upon determining that the data type of the quasi-identifier is the numeric data type and the category type of the direct identifier is the non-database identifier category, applying a binning transformation rule or a generalization transformation rule;

upon determining that the data type of the quasi-identifier is the numeric data type and the category type of the direct identifier is the database identifier category, applying a numerical offset transformation rule;

upon determining that the data type of the quasi-identifier is the string data type and the category type of the direct identifier is the non-database identifier category, applying a categorize transformation rule or a anonymize transformation rule;

upon determining that the data type of the quasi-identifier is the string data type and the category type of the direct identifier is the database identifier category, applying a format preserving encryption transformation rule;

upon determining that the data type of the quasi-identifier is the date data type and the category type of the direct identifier is the non-database identifier category, applying a binning transformation rule or a generalization transformation rule; and

upon determining that the data type of the quasi-identifier is the date data type and the category type of the direct identifier is the database identifier category, applying a date offset transformation rule.

18 . The system of claim 11 , wherein the first normalized entropy score of a data element X is calculated using:

H

⁡

(

X

)

=

-

∑

i

=

1

N

p

⁡

(

x

i

)

*

log

2

⁢

p

⁡

(

x

i

)

max

⁢

entropy

⁢

score

wherein, H(X) is the normalized entropy score of the data element X, N is a total number of possible outcomes for the data element X, p(x i ) is a probability of each of the possible outcomes of the data element X occurring, Σ denotes a sum of all possible outcomes for the data element X and the max entropy score is calculated using the equation −Σ i=1 N p(x i )*log 2 p(x i ) where p(x i ) is equal to 1/N.

19 . The system of claim 11 , wherein the second normalized entropy score of a data element X is calculated using

H

⁡

(

X

)

=

-

∑

i

=

1

N

p

⁡

(

x

i

)

*

log

2

⁢

p

⁡

(

x

i

)

max

⁢

entropy

⁢

score

wherein, H(X) is the normalized entropy score of the data element X, N is a total number of possible outcomes for the data element X, p(x i ) is a probability of each of the possible outcomes of the data element X occurring, Σ denotes a sum of all possible outcomes for the data element X and the max entropy score is calculated using the equation −Σ i=1 N p(x i )*log 2 p(x i ) where p(x i ) is equal to 1/N; and

wherein the external influence score of a data element Y is calculated using:

External

⁢

influence

⁢

score

⁢

(

Y

)

=

1

-

N

E

(

T

-

Y

)

N

E

(

T

)

wherein, N E (T−Y) is a number of unique groups that can be formed after the data element Y is removed from the non-direct identifier structured dataset; and N E (T) is the maximum number of unique groups that can exist within the non-direct identifier structured dataset.

20 . A method to de-identify personally identifiable information (PII) within a structured dataset, the method comprising:

performing a PII identification operation by:

receiving the structured dataset, the structured dataset including a plurality of data elements;

performing a direct identifier detection operation by:

identifying direct identifiers within the structured dataset based on a direct identifier threshold value; and

creating a non-direct identifier structured dataset by removing the direct identifiers from the structured dataset;

performing a quasi-identifier detection operation by:

identifying quasi-identifiers within the non-direct identifier structured dataset based on a quasi-identifier threshold value; and

creating a non-quasi-identifier structured dataset by removing the quasi-identifiers from the non-direct identifier structured dataset;

performing a unique value detection operation by:

identifying unique values within the non-quasi-identifier structured dataset based on a unique value threshold value;

performing a transformation operation by:

applying one or more transformation rules to transform each of the direct identifiers, the quasi-identifiers and the unique values into respective de-identified data elements; and

generating the de-identified structured dataset by replacing the direct identifiers, the quasi-identifiers and the unique values within the structured dataset with the respective de-identified data elements;

performing a re-identification risk assessment operation by:

evaluating a risk of re-identification of the PII within the de-identified structured dataset;

upon determining that the risk of re-identification of the PII within the de-identified structured dataset meets or exceeds a predetermined re-identification threshold value,

generating an adjusted quasi-identifier threshold values by reducing the quasi-identifier threshold value by a first predetermined amount;

generating an adjusted unique value threshold value by reducing the unique value threshold value by a second predetermined amount; and

until determining that the risk of re-identification of the PII within the de-identified structured dataset is less than the re-identification threshold value, iteratively repeating the quasi-identifier detection operation, the unique value detection operation, the transformation operation and the re-identification risk assessment operation after replacing the quasi-identifier threshold value with the adjusted quasi-identifier threshold value and replacing the unique value threshold value with the adjusted unique value threshold value.

Assignments (2)

Continuity (1)

References Cited (9)