System and method for determining when an update of an online fraud detection machine learning (ML) model is required

A computerized-system for determining when an update of an online fraud detection Machine Learning (ML) model is required, said computerized-system comprising:

• a database of labeled and unlabeled batches of financial transactions;
• a feature engineering component;
• a drift detection model;
• a trained MetaBDMM model:
• a memory to store the database of labeled and unlabeled batches;
• and
• a processor, said processor is configured to:
• (i) receive a batch of financial transactions data,
• wherein each financial transaction in the batch of financial transactions data includes one or more features,
• (ii) operate a feature engineering component to select a set of features from the one or more features of the financial transactions;
• (iii) detect a drift and a drift type in each feature in the selected set of features, by operating the drift detection model thereon;
• (iv) generate a batch-representation-vector of drift type for each feature in the selected set of features,
• wherein each feature in the selected set of features is assigned a weight:
• (vii) receive a predicted-decision of update-needed by forwarding the generated batch-representation-vector to a trained Meta-based Decision Making (MetaBDMM) model, wherein the predicted-decision of update-needed of the MetaBDMM model is based on a drift type of a preconfigured number of features in the selected set of features having highest assigned weight, and
• wherein the predicted-decision of update-needed is one of:
• (a) update-needed;
• and (b) update-not-needed, and
• (v) forward the predicted-decision of update-needed to the online fraud detection ML model,
• wherein when the predicted-decision of update-needed is update-needed, the online fraud detection ML model performs an update and when the predicted-decision of update-needed is update-not-needed the online fraud detection ML model doesn't perform the update.