What Is Feature ablation?

Feature ablation evaluates whether a feature or feature family adds incremental value by comparing an otherwise identical model with and without it. The comparison should use the same data rows, split boundaries, target definition, tuning policy, and downstream portfolio rules, so that the difference can be attributed to the removed or added input.

Ablation is stronger evidence than feature importance alone. Importance describes how a fitted model used an input among the other inputs available to it; an ablation tests whether performance deteriorates when that input is unavailable. In trading research, assess both predictive metrics and cost-aware strategy outcomes, and repeat the test across out-of-sample folds.

See also

• Feature selection

• Out-of-sample testing

• Feature engineering

Research and literature

• Isabelle Guyon and André Elisseeff, An Introduction to Variable and Feature Selection, Journal of Machine Learning Research 3, 2003.