Full control that keeps pace with your ideas.

Features · Modeling

Choose the error structure, link function, and feature shapes. Then refit, refine, and experiment quickly.

A model workspace for the Auto Pricing Frequency GLM (Poisson with a log link) on the Lift Chart tab. The main pane plots Weight bars across ten predicted quantiles with two Claim Frequency lines, Unsimplified (solid) and Simplified (dashed), both rising from the lowest to the highest quantile. A sidebar lists the model properties, filters, partitions, and features, including a driver_age spline and a vehicle_power variate.

Lift Chart

Model setup

Configure the model around your data.

Pick the math, what data goes in, and how it's grouped.

A New model form with fields for the model name, target, error structure, and link function. The error structure is set to Poisson, and the open link function dropdown shows Log, Identity, and Square Root as selectable while the link functions that are not valid for a Poisson error structure are greyed out.
New model

Automated feature creation

Generate a feature set automatically.

Choose the columns, and a gradient-boosting model derives a strong baseline of features for you.

An automated feature creation panel that starts by selecting a model (Claim Frequency Model) and its branch (Default), followed by a stack of dropdowns: variables to include (vehicle_age, bonus_malus, population_density), locked features (driver_age and vehicle_power), seeded features (vehicle_brand), filters (Train), and partitions (States), plus a Generate action.
Automated feature creation

Feature engineering

Or shape every feature by hand.

Custom factors, variates, splines, interactions, and offsets, every one tuned down to the last parameter.

A New feature panel with tabs for Custom Factor, Variate, Spline, Interaction, and Offset. The Spline tab is active for claim_count, showing a natural cubic spline with three internal knots across a range of 0 to 16, with controls to add or remove knots and switch between Natural and B-Spline.
New feature

Spline Type

Natural cubic spline with linear extrapolation at boundaries

Spline Configuration

Set the knot positions for the spline transformation.

Knots

3 internal knots

0481216
Min: 0Max: 16

Branches

Experiment without risk.

Branches keep each model fit intact. Merge the best versions through a review.

A models table listing a Claim Frequency Model (Poisson, log), a Claim Severity Model (Gamma, inverse), and a Policy Renewal Model (Binomial, logit), each on its dataset with a branch count that expands. A search field and a New model button sit above the table.

Models

Model Name
Branches
Dataset
Error Structure
Link Function

Claim Frequency Model

Poisson

Log

DefaultDefault
auto-features
spline-tuning

Claim Severity Model

Gamma

Inverse

Policy Renewal Model

Binomial

Logit

More features

Explore the rest.

Ready to ship better models?

Download Prediction Lab, or book a 15-minute demo with our team.

Download