Within which stage of the AI development life cycle should effective feature engineering be conducted?

Feature engineering belongs in the Development stage because it involves the hands-on transformation, selection, and construction of input variables that directly shape how a model learns from data. Developers iteratively create and refine features in tight feedback loops with the model training process, adjusting based on what improves accuracy and generalization. This work must occur before testing so the model is evaluated on finalized, engineered representations.

Testing is for validating model performance on held-out data, not for constructing or transforming the features the model was trained on.

Design focuses on system architecture and planning decisions, not on the technical manipulation of individual data attributes used in training.

The Define stage establishes problem scope, objectives, and success criteria - it does not involve the technical preprocessing of training data.