Computerized-method and system for predicting a probability of fraudulent financial-account access

A computerized-system for predicting a probability of fraudulent financial-account access, said computerized-system comprising:

• a memory to store a customers-account database and a persistent storage;
• and one or more processors, said one or more processors are configured to:
• a. build a Machine Learning (ML) sequence model by:
• (i) retrieving one or more chronical-sequences of a preconfigured number of non-financial activities performed on each financial-account before conducting a financial activity during a latter preconfigured period and a label of fraud or non-fraud which is associated to each compatible financial-activity, from a customers-account database;
• (ii) labeling each one or more chronical-sequences as fraud or non-fraud, based on the label of fraud or non-fraud which is associated to each compatible financial-activity that the chronical-sequence has preceded;
• (iii) providing the labeled one or more chronical-sequences to a data-vectorization model to:
• (a) encode each non-financial activity in each chronical-sequence into a unique integer value and each chronical-sequence of the one or more chronical-sequences as a vector, thus yielding an array of vectors from the labeled one or more chronical-sequences;
• and (b) generate a dictionary of vector encodings from each non-financial activity type and compatible unique integer value;
• (iv) exporting the dictionary of vector encodings to the persistent storage;
• and
• (v) training the ML sequence model to predict fraud in a chronical-sequence vector, based on a sampling, wherein the sampling is the array of vectors;
• b. implement a forward propagation routine in an encapsulated environment that runs applications to mimic a process of the ML sequence model, wherein the forward propagation routine is mimicking a process of a chronical-sequence of a preconfigured number of non-financial activities sequence vector, layer by layer to generate a fraud probability score and extracting weights and biases from each layer of the trained ML sequence model;
• and
• c. extract the extracted weights, biases and convert the forward propagation routine to an executable for integration with a Fraud Management System, in a production environment,
• wherein the Fraud Management System is operating the integrated executable to predict a probability of fraudulent financial-account access by providing a fraud probability score to each chronical-sequence of a preconfigured number of non-financial activities, provided in real-time.