Techniques for detecting emerging patterns in data used for machine learning model-based decisions

A device for detecting emerging patterns in transaction data, comprising:

• one or more memories storing instructions;
• and
• one or more processors coupled to the one or more memories and configured to execute the instructions to:
• obtain, for a collection of transaction data including multiple transactions, an embedding associated with each transaction of the multiple transactions;
• determine, for the transaction data and based on a similarity between the embeddings, neighboring sets of transactions having similar features;
• separate, based on the neighboring sets of transactions, at least a portion of the transaction data into a training dataset for training a machine learning (ML) model and a validation set for assessing performance of the ML model;
• correct the ML model at least in part by removing a portion of the multiple transactions from the training dataset;
• and
• modify, based on output of the ML model for at least a first portion of transactions in the neighboring sets of transactions, at least a second portion of transactions in the neighboring sets of transactions to be associated with a decision.

A computer-implemented method for detecting emerging patterns in transaction data, comprising:

• generating, from received transaction data, multiple patterns of the transaction data based on feature combination, where each pattern of the transaction data has a same number of common attributes;
• filtering the multiple patterns based on a parameter related to a number of data entries in each pattern to generate a filtered set of patterns;
• for each pattern in the filtered set of patterns, creating a first set of transactions associated with a decision output by a transaction service and a second set of transactions that do not result in the decision;
• deriving a confidence interval of average similarity scores based on a first set of similarity scores between transactions in the first set of transactions;
• for each transaction in the second set of transactions, determining a third set of transactions having an average similarity score that is above a lower threshold of the confidence interval;
• and
• modifying at least a portion of the third set of transactions to be associated with the decision.