EAA Web Session: Machine Learning and Anomaly Detection

Machine Learning (ML) allows computers to process data, analyse it in real time and learn and make decisions based on data. Diverse applications from self-driving cars to chess computers have successfully relied on ML.

While the insurance industry is not necessarily known for being particularly innovative, insurance companies are increasingly embracing approaches commonly used in ML to address business challenges in different areas. Actuaries and data scientists apply ML to claim management, underwriting or customer service.

Nowadays, both data and models can be processed much faster than before which means data-driven approaches to actuarial modelling are being increasingly adopted by the insurance companies. The amount of data being used by insurance companies for different purposes has increased exponentially. As such, it is becoming more difficult for actuaries to identify anomalies in data, models and outputs. For example, some insurers apply Least Square Monte Carlo methods to derive their Net Asset Value and Best Estimate Liability proxy models. These models take a huge amount of data to perform complex calculations. It is not possible for actuaries to understand bad data and model behaviour by using traditional methods given the amount of data involved.

In this web session, we are going to discuss how techniques commonly used by data scientist in ML applications can help actuaries detect/remove bad data and significantly improve forecasting abilities of modern actuarial models.

Registration deadline: 30 March 2022