Method and system for detecting fraudulent transactions using a fraud detection model trained based on dynamic time segments

A system, comprising:

• a memory having instructions executable stored thereon;
• and
• a processor configured to execute the executable instructions in order to perform an operation for detecting fraudulent transactions in a transaction processing system:
• receive a request to process a transaction;
• generating an input data set including a vector representing the transaction and a plurality of vectors representing historical transactions;
• dividing the input data set into a plurality of ragged tensors, each ragged tensor corresponding to a non-overlapping time segment of variable length and having a plurality of vectors associated with dates within the time segment, and the non-overlapping time segment of variable length being based on an amount of time prior to a time associated with the transaction such that a ragged tensor associated with historical transactions having timestamps closer to the time associated with the transaction covers a smaller amount of time than a ragged tensor associated with historical transactions having timestamps further from the time associated with the transaction;
• generate a reduced input data set by generating, for each respective ragged tensor of the plurality of ragged tensors, a respective representative vector using max pooling over vectors in the ragged tensor;
• generate, using a fraud detection model, a fraudulent transaction score based on the reduced input data set, wherein the fraud detection model comprises a machine learning model trained to predict a likelihood of a transaction being a fraudulent transaction based on representative tensors representing transactions recorded in non-overlapping variable length time windows;
• and
• process the transaction based, at least in part, on the fraudulent transaction score, wherein in order to process the transaction, the processor is configured to:
• determine that the transaction is likely to be a fraudulent transaction based on a comparison of the fraudulent transaction score to a threshold score;
• complete the transaction;
• and
• block a user from taking action with respect to the transaction until an amount of time has elapsed from the completing the transaction.