| CPC G06Q 20/401 (2013.01) [G06Q 20/04 (2013.01); G06Q 20/12 (2013.01); G06Q 30/0631 (2013.01)] | 20 Claims |
|
1. A server system, comprising:
a memory comprising executable instructions; and
a processor operationally coupled to the memory, the processor configured to:
access historical transaction data associated with a merchant, the historical transaction data including transaction-level data of past payment transaction requests and product information which were applied to the past payment transaction requests;
aggregate the past payment transaction requests according to specific data elements;
obtain past payment transaction features based on the aggregated past payment transaction requests;
receive, by a data pre-processing engine, a payment authorization request for a payment transaction between a cardholder and a merchant in real time, and identify payment transaction features associated with the payment transaction based, at least in part, on the payment authorization request;
predict, by a reinforcement learning (RL) agent, a combination of one or more authorizing components to be applied to the payment transaction to obtain an authorization decision product recommendation strategy for the payment transaction, the combination of the one or more authorizing components predicted based, at least in part, on application of a trained deep RL model to the payment transaction features,
wherein the deep RL model is trained based, at least, on the obtained past payment transaction features and authorization decision products applied to the past payment transaction requests by:
defining a state space of the deep RL model that includes a plurality of states,
defining an action space of the deep RL model that includes a plurality of actions,
calculating a reward value that is obtained in any state using a reward function, the reward function representing a reward value of a state after performing an action, the reward function further being a function of (i) a probability of approval of a transaction, (ii) a probability of the transaction being fraud (iii) a cost incurred on applying a product on the transaction and (iv) a summation of total cost of all products that are applied to the payment transaction, wherein the probability of approval of a transaction and the probability of the transaction being fraud are determined based on the obtained payment transaction features, and
transmit the payment authorization request and the obtained authorization decision product recommendation strategy to an issuer associated with the cardholder.
|