Method for extreme class imbalance within fraud detection

Abstract

A computerized-method for building ensemble of supervised and unsupervised Machine Learning (ML) models for fraud-predictions, for a client having an extremely-imbalanced-dataset, is provided herein. The computerized-method includes: (i) receiving an extremely-imbalanced-dataset from a client for building a ML model; (ii) retrieving datasets of other clients; (iii) identifying a rate-of-dataset-imbalance for each retrieved dataset; (iv) routing each dataset of 'K' datasets with identified rate-of-dataset-imbalance above a preconfigured-threshold to supervised ML models for training thereof and to yield a trained-object; (v) training a meta-learning-supervised ML model by providing the ‘K’ yielded trained-objects; (vi) routing each dataset of ‘L’ datasets with identified rate-of-dataset-imbalance below a preconfigured-threshold to an unsupervised ML model to generate clusters; (vii) combining the ‘k’ supervised ML models and the ‘L’ unsupervised ML models into ensemble ML models; and (viii) deploying the ensemble ML models in a financial-system in production-environment for prediction of fraud in a financial-transaction.