Automated Feature Engineering for Rating Models

Purpose

Automated Feature Engineering for Rating Models is designed to automatically identify and engineer new predictive variables from raw data using machine learning and data mining, with a focus on supporting rating model development in an insurance context.

Primary users

The primary user is not specified in the provided information. The agent is associated with the AQS team and is intended for users involved in working with insurance data and rating model feature development.

Where it fits (process/stage/trigger)

This agent fits into the feature discovery and preparation stage of rating model development, particularly when raw insurance-related datasets are available and need to be explored for potential predictive variables.

Key capabilities / workflow

The workflow analyzes raw data, checks whether sufficient information is available, extracts candidate variables, evaluates whether predictive signal is present, engineers new features, validates them for rating models, and delivers the engineered variables for further use.

Inputs

Inputs are not specified beyond the available dataset types, which include policy-level data, claims, credit scores, demographics, and telematics data.

Outputs / Deliverables

Outputs are not specified beyond the use case objective of identifying and engineering new predictive variables from raw data for rating models.

Value

The value of the agent is to support automated discovery and creation of predictive variables, helping rating model teams make better use of available insurance datasets while reducing the manual effort required for feature engineering.