The IBM Developer JumpStart program provides professional development, networking and skills enhancement opportunities for our entry-level software engineers. As a part of IBM Developer JumpStart, IMS proposed an idea of using machine learning with IMS data. Eight new developers worked on this exciting project. The main goal was to demonstrate that with the help of new technology like machine learning, you can add value to your existing IMS assets.
At a high level, machine learning is a way of building scoring models based on your data. Many IMS customers are in the financial industry; the data they keep in IMS is frequently considered their operational data.
One of the use cases for our project illustrates how banks go through their data to search for fraudulent activity. This might take a lot of time, and throughout most of the scenario, fraudulent transactions were detected after the processing of transactions, which is too late. This is not an efficient way of detecting fraudulent transactions. What if you can detect this activity at the point of occurrence, when the transaction is being processed, prior to payment processing? Early detection can lead to a decrease in the amount of money that the bank has to pay out to fraud victims; it can also decrease the amount of manual work of finding fraudulent transactions, notify users regarding frauds, send out new cards etc.
Fraud analysts help to identify those fraudulent transactions; the task of identifying those transactions can require a lot of manual work and paperwork. We wondered, â€œCould machine learning provide a better way to detect fraudulent transactions?” Our main goal for the project was to design a system that could automatically detect the fraudulent transactions so that analysts do not have to manually determine whether transactions are fraudulent. This would result in less manual work and a decrease in false positives.
Building a scoring model with IMS data
To identify fraudulent transactions, we need to design the scoring service that will predict the likelihood of fraudulent transactions. IBM provides tools that you can use to build scoring services based on your needs. One is IBM Watson Studio, which can be used to build, train, deploy and manage scoring models. Another tool is IBM Machine Learning for z/OS (MLz).
IBM makes it possible through MLz for mainframe users to keep their enterprise data on the mainframe while building out their scoring models. MLz allows your potentially sensitive data to stay on a secured environment: the mainframe. MLz also has the added benefit of avoiding potentially costly extract, transform and load (ETL) operations related to moving data. For this project, we used IBM Watson Studio. We tested different algorithms and chose the model that gave the highest accuracy, around 82%. The model that we decided to use is Random Forest, the most simple and widely used algorithm in machine learning. Random Forest builds multiple decision trees and merges them to create a large decision tree that gives you more accurate and stable predictions.
Invoking a scoring model from a cloud application
With the help of IBM Watson Studio, you can deploy your scoring models as RESTful services. We created a front end application on IBM Cloud that invoked the machine learning scoring service through the RESTFUL service and displayed the results of the scoring model that will show likelihood of false transactions.
Take a look at these tools for machine learning to learn more about it!