What Is Feature?

In machine learning, a feature is an input variable used by a model to make a prediction. In a tabular dataset, features are usually the columns of the input matrix; in quantitative finance they may include lagged returns, accounting characteristics, market data, macroeconomic variables, or alternative data.

A feature must satisfy decision-time availability. A variable that is only known after the predicted outcome, or whose historical value has been revised after the fact, is a source of data leakage and can make a backtest appear predictive when it is not. Feature engineering creates or transforms features; it does not remove the need for point-in-time data and out-of-sample testing.

See also

• Label

• Target variable

• Look-ahead bias

• Feature engineering

• Feature selection

Research and literature

• scikit-learn glossary: feature.

• Shihao Gu, Bryan Kelly, and Dacheng Xiu, Empirical Asset Pricing via Machine Learning, The Review of Financial Studies 33(5), 2020.