Fraud detection methods and systems based on evolution-based black-box attack models

A server system configured for generation of adversarial samples for machine learning model training on tabular data, comprising:

• a communication interface;
• a memory comprising executable instructions;
• and
• a processor communicably coupled to the communication interface and the memory, the processor configured to execute the instructions to cause the server system, at least in part, to:
• access tabular dataset samples;
• initialize a plurality of encoders, wherein weights of each of the plurality of encoders are randomly initialized;
• compute a set of initial adversarial samples using the plurality of encoders based, at least in part, on the tabular dataset samples, wherein the set of initial adversarial samples simulates perturbations in the tabular dataset samples;
• generate a plurality of evolved encoders based, at least in part, on adjusting the weights of the plurality of encoders based on feedback from a fitness scoring function;
• compute a plurality of fitness scores for the plurality of evolved encoders;
• select a top-performing evolved encoder of the plurality of evolved encoders based on the plurality of fitness scores;
• and
• generate a set of final adversarial samples using the top-performing evolved encoder based, at least in part, on the tabular dataset samples, wherein the set of final adversarial samples is used to train a machine learning model.

A non-transitory computer-readable storage medium comprising computer-executable instructions that, when executed by at least a processor of a server system, cause the server system to perform a method for generation of adversarial samples for machine learning model training on tabular data, the method comprising:

• accessing tabular dataset samples;
• initializing a plurality of encoders, wherein weights of each of the plurality of encoders are randomly initialized;
• computing a set of initial adversarial samples using the plurality of encoders based, at least in part, on the tabular dataset samples, wherein the set of initial adversarial samples simulates perturbations in the tabular dataset samples;
• generating a plurality of evolved encoders, based, at least in part, on adjusting the weights of the plurality of encoders based on feedback from a fitness scoring function;
• computing a plurality of fitness scores for the plurality of evolved encoders;
• selecting, by the server system, a top-performing evolved encoder of the plurality of evolved encoders based on the plurality of fitness scores;
• and
• generating a set of final adversarial samples using the top-performing evolved encoder based, at least in part, on the tabular dataset samples, wherein the set of final adversarial samples is used to train a machine learning model.