In CRISP-DM, if a problem is detected at the Modeling step, you should return to which step?

• A. Business understanding
• B. Data preparation
• C. Deployment
• D. Data collection

Answer: (B)

When modeling produces poor results, the fix usually lies in how the data is prepared. The modeling stage relies on the quality and representation of input data, so if performance is lacking, you typically return to the data preparation phase to clean the data, engineer or create better features, handle missing values, encode categorical variables appropriately, normalize or scale features, and address issues like data leakage or class imbalance. By improving the inputs, the model has a fair chance to learn meaningful patterns, and you can re-run modeling to see if performance improves.

Revisiting business goals would be about redefining objectives, not fixing data inputs; deployment is the final stage after a satisfactory model is achieved; collecting more data could be considered later, but the most immediate adjustment for modeling problems is to refine how the data is prepared.