2. Machine Learning - An Overview

2.1 Introduction

Machine Learning is heavily dependent on historical data to build models and use the models to predict outcomes for better decision making

The following points encompasses contextual information to help you get familiarized with the subject.

• Machine learning model building is dependent on historical data, spread across dates
• Model building will take time. It may range from couple of minutes to hours, sometime days, depending on the algorithm complexity
• Model has to be built for each bank. It cannot be pre-shipped not it can be re-used across banks.
• Model is ONLY as good as the data used to build it
• Built Model are valid only for the data used to build the model
• The pre-shipped ORE code is designed to work with the data model provided as part of the release
• The modelling code may need to be modified under the following circumstances
  • The underlying data limits model performance and mandates modifying the
    pre-shipped ORE code
  • Data Model is altered, on business demand

This chapter contains the following section:

• Section 2.2, "Machine Learning in Oracle FLEXCUBE"
• Section 2.3, "Pre-requisites"

2.2 Machine Learning in Oracle FLEXCUBE

The machine learning framework is embedded within Oracle FLEXCUBE. It is built using Oracle R Enterprise, which is separately licensed.

The business cases are limited to retail customers.

2.3 Pre-requisites

The following Pre-requisites are mandatory for the Machine Learning Framework.

Environmental

 

• Separate instance of Machine Learning Server.

Note

ORE database MUST be a separate instance from Oracle FLEXCUBE Instance. This is to ensure that the machine learning algorithmic processing workload is not on the critical path of banking operations and consume critical FLEXCUBE server memory

• Machine Learning Database is Oracle 18c
• Oracle R Enterprise is installed in the Machine Learning Server

Historical Data

 

• Date wise transaction history in Oracle FLEXCUBE is mandatorily required for building machine learning models.
• Consulting/Solution team to consider extracting legacy transaction history as part of
  migration strategy

Oracle FLEXCUBE Data Replication

 

• Data replication from FLEXCUBE to the Machine learning server is outside the purview of this manual.
• Any data replication software could be deployed to replicate Oracle FLEXCUBE data to the Machine Learning database Instance
• It is left to the bank/solution team to decide on the data Replication Software including the mode of replication and frequency of replication

Refer to the Installation Guide for completing the installation, prior to initiating the model building routine.