Configure Your Model

Target Variable

Select the column you want to predict (e.g., claim frequency, loss amount, retention).

Objective Function

Choose the objective function that matches your target variable's statistical distribution:

Objective	Use Case	Target Type	Link Function	Factor Tables
Poisson	Claim counts, frequency	Non-negative integers	Log	Multiplicative
Gamma	Claim severity, loss amounts	Positive continuous	Log	Multiplicative
Tweedie	Pure premium, aggregate loss	Zero-inflated positive	Log	Multiplicative
Binary	Conversion, retention	0 or 1	Logistic	Additive
Regression	General continuous	Any continuous	Identity	Additive
Huber	Robust regression	Continuous with outliers	Identity	Additive

Important: The objective function determines both the loss function used during training and the link function for the final model.

Feature Selection

Select columns to use as predictors:

• Include all potentially relevant variables
• Exclude identifiers (e.g., policy ID, customer ID)
• Exclude leakage variables (information not available at prediction time)

The GBM will automatically perform feature selection during training—features that don't improve predictions will be excluded from the final model.

Before Starting Training

Verify:

• Objective function matches your target's distribution
• Feature list includes all relevant predictors
• No data quality warnings displayed

Once configuration is complete, training begins automatically.

Next Steps

After training completes: Select a model from the Pareto frontier