Method for adaptive tuning via automated simulation and optimization

1. A computer implemented method for optimization of model parameters of at least one predictive model for detecting suspicious financial activity, the predictive model for determining a risk score and comprising functions, each function having a plurality of key indicators and a score corresponding to each of the key indicators, each of the key indicators being a range of values corresponding to its function, the method comprising:

receiving, by a processor, the at least one predictive model;

receiving, by the processor, information regarding financial transaction data, financial account profile data, fraud alert feedbacks, or any combination thereof;

determining, by the processor, an influence ranking for each key indicator in the at least one predictive model based on the received information and based on sensitivity analysis techniques;

selecting, by the processor, a reduced set of key indicators and corresponding scores to optimize from each of the at least one predictive model, each key indicator and corresponding score in the reduced set having an influence ranking above a predetermined influence ranking;

evaluating, by the processor, each reduced set of key indicators and corresponding scores using quality metrics;

based on the evaluation of each reduced set, selecting, by the processor, a best performing model candidate;

generating, by the processor, at least one random model, the random model having random sets of scores for key indicators of the best performing model candidate;

performing, by the processor, gradient-ascent functional optimization on the best performing model candidate and the at least one random model to generate a set of at least two new models for each of the best performing model candidate and the at least one random model;

calculating, by the processor, a performance ranking of each new model in the set of new models; and

selecting, by the processor, an optimized model, wherein the optimized model is the new model with the highest performance ranking.

2. The method of claim 1 , further comprising determining a level of improvement between the optimized model and the best performing model candidate.

3. The method of claim 2 , wherein,

if the level of improvement between the optimized model and the best performing model candidate is above a predetermined threshold, iteratively:

generating at least one new random model, the at least one new random model having random sets of scores for key indicators of the optimized model,

performing, by the processor, gradient-ascent functional optimization on the optimized model and the at least one random model to generate a set of at least two additional models for each of the optimized model and the at least one new random model,

calculating, by the processor, a performance ranking of each additional model in the set of additional models,

selecting, by the processor, a new optimized model, wherein the new optimized model is the additional model with the highest performance ranking,

determining a level of improvement between the additional model with the highest performance ranking and the optimized model, and

replacing the scores for the optimized model with the scores of the new optimized model if the new optimized model is an improvement of the optimized model; and

if the level of improvement between the optimized model and the best performing model candidate is equal to or below a predetermined threshold, selecting the optimized model for detecting suspicious financial activity.

4. The method of claim 1 , further comprising, selecting at least one relevant model from the received at least one predictive model based on a type of a financial account from which funds are to be transferred from, the amount of money being transferred, or any combination thereof.

5. The method of claim 4 , wherein

the influence ranking is determined for each key indicator in the at least one relevant model based on the received information and based on the sensitivity analysis techniques; and

the reduced set of key indicators and corresponding scores to optimize are selected from the at least one relevant model.

6. The method of claim 1 , wherein selecting a reduced set of key indicators and corresponding scores further comprises:

for each key indicator and corresponding score in the at least one predictive model, determining, by the processor, if there is another key indicator and corresponding score in the at least one predictive model which forms a pair of key indicators with a Pearson value over a predetermined Pearson value; and

for each pair of key indicators with a Pearson value over the predetermined Pearson value, removing, by the processor, the key indicator and corresponding score of the key indicator with lowest influence ranking in the pair of key indicators with a Pearson value over the predetermined Pearson value from the reduced set of key indicators.

7. The method of claim 6 , wherein the predetermined Pearson value is approximately 0.7.

8. The method of claim 1 , further comprising:

determining the risk score of a transaction by adding together each score for each key indicator of the reduced set of key indicators in the optimized model; and

if the risk score of the transaction is above a predetermined risk score:

notifying an account holder of the account from which the transaction originated,

blocking the transaction,

requesting authentication from the account holder, or

any combination thereof.

9. The method of claim 1 , wherein:

receiving the at least one predictive model comprises receiving, by the processor, a plurality of predictive models;

selecting the best performing model candidate comprises selecting, by the processor, more than one best performing model candidates based on the evaluation of each reduced set; and

the gradient-ascent functional optimization is performed, by the processor, on the more than one best performing model candidates and the at least one random model to generate a set of at least two new models for each of the best performing model candidates and the at least one random model.

10. A system for optimization of model parameters of at least one predictive model for detecting suspicious financial activity, the predictive model for determining a risk score and comprising functions, each function having a plurality of key indicators and a score corresponding to each of the key indicators, each of the key indicators being a range of values corresponding to its function, the system comprising:

a server, and

a processor configured to:

receive the at least one predictive model,

receive information regarding financial transaction data, financial account profile data, fraud alert feedbacks, or any combination thereof from the server,

determine an influence ranking for each key indicator in the at least one predictive model based on the received information and based on sensitivity analysis techniques,

select a reduced set of key indicators and corresponding scores to optimize from each of the at least one predictive model, each key indicator and corresponding score in the reduced set having an influence ranking above a predetermined influence ranking,

evaluate each reduced set of key indicators and corresponding scores using quality metrics,

select a best performing model candidate based on the evaluation of each reduced set,

generate at least one random model, the random model having random sets of scores for key indicators of the best performing model candidate,

perform gradient-ascent functional optimization on the best performing model candidate and the at least one random model to generate a set of at least two new models for each of the best performing model candidate and the at least one random model,

calculate a performance ranking of each new model in the set of new models, and

select an optimized model, wherein the optimized model is the new model with the highest performance ranking.

11. The system of claim 10 , wherein the processor is configured to determine a level of improvement between the optimized model and the best performing model candidate.

12. The system of claim 11 , wherein the processor is configured to:

if the level of improvement between the optimized model and the best performing model candidate is above a predetermined threshold, iteratively:

generate at least one new random model, the at least one new random model having random sets of scores for key indicators of the optimized model;

perform gradient-ascent functional optimization on the optimized model and the at least one random model to generate a set of at least two additional models for each of the optimized model and the at least one new random model;

calculate a performance ranking of each additional model in the set of additional models;

select a new optimized model, wherein the new optimized model is the additional model with the highest performance ranking;

determine a level of improvement between the additional model with the highest performance ranking and the optimized model;

replace the scores for the optimized model with the scores of the new optimized model if the new optimized model is an improvement of the optimized model; and

if the level of improvement between the optimized model and the best performing model candidate is equal to or below a predetermined threshold, select the optimized model for detecting suspicious financial activity.

13. The system of claim 10 , wherein the processor is configured to select at least one relevant model from the received at least one predictive model based on a type of a financial account from which funds are to be transferred from, the amount of money being transferred, or any combination thereof.

14. The system of claim 13 , wherein the processor is configured to:

determine the influence ranking for each key indicator in the at least one relevant model based on the received information and based on the sensitivity analysis techniques; and

select the reduced set of key indicators and corresponding scores from the at least one relevant model to optimize.

15. The system of claim 10 , wherein the processor is configured to:

for each key indicator and corresponding score in the at least one predictive model, determine if there is another key indicator and corresponding score in the at least one predictive model which forms a pair of key indicators with a Pearson value over a predetermined Pearson value; and

for each pair of key indicators with a Pearson value over the predetermined Pearson value, remove the key indicator and corresponding score of the key indicator with lowest influence ranking in the pair of key indicators with a Pearson value over the predetermined Pearson value from the reduced set of key indicators.

16. The system of claim 15 , wherein the predetermined Pearson value is approximately 0.7.

17. The system of claim 10 , wherein the processor is configured to:

determine the risk score of a transaction by adding together each score for each key indicator of the reduced set of key indicators in the optimized model; and

if the risk score of the transaction is above a predetermined risk score:

notify an account holder of the account from which the transaction originated,

block the transaction,

request authentication from the account holder, or

any combination thereof.

18. The system of claim 10 , wherein the processor is configured to:

receive a plurality of predictive models;

select more than one best performing model candidates based on the evaluation of each reduced set; and

perform gradient-ascent functional optimization on the more than one best performing model candidates and the at least one random model to generate a set of at least two new models for each of the best performing model candidates and the at least one random model.