Explaining machine learning models by tracked behavioral latent features

1. A method comprising:

receiving, by one or more computer processors, a time series of input variables representing a time-ordered sequence of Observations associated with an entity;

generating, by the one or more computer processors, a word dictionary for the input variables, the word dictionary representing a unique sequence of event words across the entity and a group of other entities;

generating, by the one or more computer processors, a set of archetypes from the unique sequence of event words, where topic modeling associates a loading of archetypes associated with each event word;

generating, with a topic modeling module executed by the one or more computer processors, a vector of weights representing a distribution of the time series of input variables aligning with each of a given number of latent topic variables, the given number of topic variables corresponding to latent features of behavior based on variable values;

producing, according to a machine learning model executed by the one or more computer processors, a detection score associated with the entity for each input event in the time-ordered sequence of observations;

generating, by the one or more computer processors, one or more explanations, each explanation describing a reason corresponding to the produced detection score;

filtering, through comparison against profiled average values of latent topic variables by the one or more computer processors, the one or more explanations to generate a filtered list of explanations according to archetype, the filtered list providing a list of archetypes associated with the detection score and increasing in relevance of behavior; and

ranking, according to a closeness of a bin of historical latent topic variables' expected score to the current score on the event, the filtered list of archetypes and associated reasons that most affect the detection score.

2. The method in accordance with claim 1 , the machine learning model includes a neural network.

3. The method in accordance with claim 1 , wherein the ranking of the filtered list of archetypes that most affect the detection score according to a closeness of a bin of historical latent topic variables' expected score to the current score on the event is evaluated by one or more computer processors and a precomputed lookup table.

4. The method in accordance with claim 3 , further comprising:

compiling, by the one or more computer processors, the lookup table to estimate an expected score based on topic allocation by evaluating the topic allocation and score outputs of the machine learning model.

5. The method in accordance with claim 4 , further comprising:

collecting, by the one or more computer processors, statistics on a holdout dataset of historical observations that is a representative subpopulation of a complete dataset of the time series of input variables.

6. The method in accordance with claim 5 , wherein the representative subpopulation of the complete dataset includes a dataset of typical behavioral scenarios, with each typical behavioral scenario being represented by a vector of archetype values and a corresponding score.

7. A computer program product comprising a non-transitory machine-readable medium storing instructions that, when executed by at least one programmable processor, cause the at least one programmable processor to perform operations comprising:

receive a time series of input variables representing a time-ordered sequence of observations associated with an entity;

generate a word dictionary for the input variables, the word dictionary representing a unique sequence of event words across the entity and a group of other entities;

generate a set of archetypes from the unique sequence of event words, where topic modeling associates a loading of archetypes associated with each event word;

generate, with a topic modeling module, a vector of weights representing a distribution of the time series of input variables aligning with each of a given number of latent topic variables, the given number of topic variables corresponding to latent features of behavior based on variable values;

produce, according to a machine learning model, a detection score associated with the entity for each input event in the time-ordered sequence of observations;

generate one or more explanations, each explanation describing a reason corresponding to the produced detection score;

filter, through comparison against profiled average values of latent topic variables, the one or more explanations to generate a filtered list of explanations according to archetype, the filtered list providing a list of archetypes associated with the detection score and increasing in relevance of behavior; and

rank, according to a closeness of a bin of historical latent topic variables' expected score to the current score on the event, the filtered list of archetypes and associated reasons that most affect the detection score.

8. The computer program product in accordance with claim 7 , wherein the machine learning model includes a neural network.

9. The computer program product in accordance with claim 7 , wherein the ranking of the filtered list of archetypes that most affect the detection score according to a closeness of a bin of historical latent topic variables' expected score to the current score on the event is evaluated by one or more computer processors and a precomputed lookup table.

10. The computer program product in accordance with claim 9 , further comprising:

compiling, by the one or more computer processors, the lookup table to estimate an expected score based on topic allocation by evaluating the topic allocation and score outputs of the machine learning model.

11. The computer program product in accordance with claim 10 , further comprising:

collecting, by the one or more computer processors, statistics on a holdout dataset of historical observations that is a representative subpopulation of a complete dataset of the time series of input variables.

12. The computer program product in accordance with claim 11 , wherein the representative subpopulation of the complete dataset includes a dataset of typical behavioral scenarios, with each typical behavioral scenario being represented by a vector of archetype values and a corresponding score.

13. A system comprising:

a programmable processor; and

a non-transitory, machine-readable medium storing instructions that, when executed by the programmable processor, cause the at least one programmable processor to perform operations comprising:

receive a time series of input variables representing a time-ordered sequence of observations associated with an entity;

generate a word dictionary for the input variables, the word dictionary representing a unique sequence of event words across the entity and a group of other entities;

generate a set of archetypes from the unique sequence of event words, where topic modeling associates a loading of archetypes associated with each event word;

generate, with a topic modeling module, a vector of weights representing a distribution of the time series of input variables aligning with each of a given number of latent topic variables, the given number of topic variables corresponding to latent features of behavior based on variable values;

produce, according to a machine learning model, a detection score associated with the entity for each input event in the time-ordered sequence of observations;

generate one or more explanations, each explanation describing a reason corresponding to the produced detection score;

filter, through comparison against profiled average values of latent topic variables, the one or more explanations to generate a filtered list of explanations according to archetype, the filtered list providing a list of archetypes associated with the detection score and increasing in relevance of behavior; and

rank, according to a closeness of a bin of historical latent topic variables' expected score to the current score on the event, the filtered list of archetypes and associated reasons that most affect the detection score.

14. The system in accordance with claim 13 , wherein the machine learning model. includes a neural network.

15. The system in accordance with claim 13 , wherein the ranking of the filtered list of archetypes that most affect the detection score according to a closeness of a bin of historical latent topic variables' expected score to the current score on the event is evaluated by one or more computer processors and a precomputed lookup table.

16. The system in accordance with claim 15 , further comprising:

compiling, by the one or more computer processors, the lookup table to estimate an expected score based on topic allocation by evaluating the topic allocation and score outputs of the machine learning model.

17. The system in accordance with claim 16 , further comprising:

collecting, by the one or more computer processors, statistics on a holdout dataset of historical observations that is a representative subpopulation of a complete dataset of the time series of input variables.

18. The system in accordance with claim 17 , wherein the representative subpopulation of the complete dataset includes a dataset of typical behavioral scenarios, with each typical behavioral scenario being represented by a vector of archetype values and a corresponding score.