System for reducing transaction failure

1 . A method, comprising:

accessing a first plurality of card identifiers associated with a first plurality of payment instruments, respectively;

accessing a feature set that includes transaction information associated with a second plurality of card identifiers associated with a second plurality of payment instruments, wherein the transaction information comprises: for each card identifier of the second plurality of card identifiers, a quantity of declined transactions using the card identifier within a defined time period, an average transaction amount for the declined transactions, or a geographic location in which the declined transactions were performed;

training a machine learning model using the feature set as training data;

calculating, at least in part using the trained machine learning model, a decline probability score for each card identifier of the first plurality of card identifiers, wherein the decline probability score indicates a probability that the payment instrument associated with the card identifier of the first plurality of card identifiers will be declined for payment, and wherein the decline probability score for each card identifier of the first plurality of card identifiers is calculated at least in part by:

generating, at least in part via a hashing function, a plurality of hashes for the first plurality of card identifiers; and

calculating the decline probability score based on whether at least one of the plurality of hashes meets a specified criterion;

comparing, for at least a first card identifier of the first plurality of card identifiers, the calculated decline probability score with a specified threshold;

determining, based on a result of the comparing, whether to process a transaction involving the first card identifier;

sending, based on the result indicating that the calculated decline probability score meets the specified threshold, an authorization request for the first card identifier; and

terminating, based on a received response to the authorization request indicating that the authorization request was unsuccessful, the transaction involving the first card identifier.

2 . The method of claim 1 , wherein at least some of the first plurality of card identifiers are also in the second plurality of card identifiers.

3 . The method of claim 1 , further comprising removing, based on the received response to the authorization request indicating that the authorization request was unsuccessful, the payment instrument associated with the first card identifier as a payment option from a user account.

4 . The method of claim 1 , wherein the authorization request comprises a zero-dollar authorization request.

5 . The method of claim 1 , wherein the machine learning model comprises a gradient boosted tree model or a random forest model.

6 . The method of claim 1 , wherein:

the generated plurality of hashes include different sets of duplicate hashes; and

a different decline probability score is calculated for each set of the duplicate hashes.

7 . The method of claim 1 , wherein the hashing function is applied to each card identifier of the first plurality of card identifiers.

8 . The method of claim 1 , wherein the machine learning model is updated on a periodic basis.

9 . The method of claim 1 , wherein the authorization request is sent in real-time as a part of a real-time transaction processing based on the calculated decline probability score.

10 . A system, comprising:

a non-transitory memory storing instructions; and

one or more hardware processors configured to execute the instructions to cause the system to perform operations comprising:

accessing a database on which a plurality of card identifiers is stored, wherein the plurality of card identifiers is associated with a plurality of payment instruments, respectively;

accessing a gradient boosted tree model or a random forest model trained as a machine learning model, wherein training data used to train the gradient boosted tree model or the random forest model comprises: a quantity of a plurality of previous transactions that have been declined within a defined time period, an average transaction amount associated with the plurality of previous transactions that have been declined, or a geographic location associated with the plurality of previous transaction that have been declined;

determining, at least in part using the machine learning model that comprises the gradient boosted tree model or the random forest model, a score for each card identifier of the plurality of card identifiers, wherein the score represents a likelihood that the payment instrument associated with the card identifier will be declined for payment, and wherein the score for each of the card identifiers is determined at least in part by:

applying a hashing function to the plurality of the card identifiers, thereby generating a plurality of hashes; and

determining whether the plurality of hashes meets a specified criterion;

determining that the score for a first card identifier of the plurality of card identifiers meets a specified threshold;

transmitting, based on the determining that the score for the first card identifier of the plurality of card identifiers meets the specified threshold, an authorization request for the first card identifier;

receiving a response that indicates the authorization request was denied; and

terminating, based on the received response, a prospective transaction involving the first card identifier.

11 . The system of claim 10 , wherein the operations further comprise removing the payment instrument associated with the first card identifier as a payment option from a user account.

12 . The system of claim 10 , wherein:

different sets of duplicate hashes are generated by the applying of the hashing function; and

a respective score is calculated for each set of the duplicate hashes.

13 . The system of claim 10 , wherein a same hashing function is applied to each card identifier of the plurality of card identifiers.

14 . The system of claim 10 , wherein the authorization request comprises a zero-dollar authorization request.

15 . A non-transitory machine-readable medium having instructions stored thereon, wherein the instructions are executable to cause a machine of a system to perform operations comprising:

accessing a payment request involving a first card identifier that is associated with a payment instrument of a user account;

training a machine learning model using a feature set that includes transaction information associated with a plurality of other card identifiers, wherein the transaction information comprises: for each card identifier of the plurality of other card identifiers, a quantity of declined transactions using the other card identifier within a defined time period, an average transaction amount for the declined transactions, or a geographic location in which the declined transactions were performed;

determining, at least in part using the trained machine learning model, a likelihood that the payment instrument associated with the first card identifier will be declined for payment of a transaction conductive via the user account, and wherein the determining comprises:

generating a plurality of hashes at least in part by applying a hashing function to the first card identifier;

determining a first hash of the plurality of hashes that meets a predefined criterion; and

determining, for the first hash, a score that represents the likelihood that the payment instrument associated with the first card identifier will be declined for payment;

determining, based on the determined likelihood, whether to process the payment request; and

declining, based on the determined likelihood meeting a specified threshold, the payment request.

16 . The non-transitory machine-readable medium of claim 15 , wherein the operations further comprise:

removing the payment instrument associated with the first card identifier from the user account.

17 . The non-transitory machine-readable medium of claim 15 , wherein:

the plurality of hashes include different sets of duplicate hashes; and

a different score is determined for each set of the duplicate hashes.

18 . The non-transitory machine-readable medium of claim 15 , wherein the machine learning model comprises a gradient boosted tree model or a random forest model.

19 . The non-transitory machine-readable medium of claim 15 , wherein the operations further comprise: updating the machine learning model on a periodic basis.

20 . The non-transitory machine-readable medium of claim 15 , wherein the system is a system of a service provider, and wherein the operations further comprise storing the score in a database of the service provider.