How Machine Learning Streamlines Risk Management

It is essential for us to establish the rigorous governance processes and policies that can quickly identify when the model begins to fail, said Abhaya K Srivastava, SVP at Northern Trust Corporation, at Machine Learning Developers Summit 2021. In the first talk of Day 1, Srivastava delved into how different sectors including financial, healthcare, retail etc use  emerging technologies like AI and machine learning. He talked about the use of machine learning in risk management, such as the techniques used in risk analytics across risk categories (credit/market/operational and PPNR), regulatory view on machine learning models as well as challenges for banks using ML.

The second part of the session covered related governance and policies in machine learning models. Srivastava stated, “The terms of AI are not new but businesses and organisations have started using these technologies in a different way. We have noticed the influence of machine learning in business applications, ML is playing an important role in risk management and there has been a constant focus on how risks are being detected, reported, managed, etc. Specifically, if you look at the model building, validation, audit and governance, the focus is a lot more.”

He added, “There is a lot of pressure from the regulators and senior management to predict with accuracy, but by following the regulations and policies”

The speaker mentioned that a large number of areas in risk management have significantly benefited from machine learning techniques.

Srivastava then deliberated on the techniques of machine learning. He said machine learning algorithms like neural networks, support vector machines and random forests are widely used in simple analytical tasks. He also explained how to create a machine learning model that provides maximum accuracy and is robust in nature.  

The speaker concluded the talk by discussing the importance of regulatory policies and guidelines for machine learning models. Srivastava said, “We need model governance because machine learning models learn and enable better accuracy as well as the ability to predict, but there involves an increase in the risk when someone uses a different source of data.”