Which of the following data management techniques BEST improves an AI model's performance by enhancing training data quality?

Data classification organizes data into categories for access control and policy purposes but does not improve the accuracy or correctness of the data content used in training.

Data access controls govern who can read or modify data for security purposes but have no effect on the quality or accuracy of the data itself.

Data scrubbing is the best technique for improving AI model performance through training data quality because it identifies and corrects or removes inaccurate records, duplicate entries, null values, and inconsistencies that introduce noise into the learning process. A model trained on clean, accurate data develops more reliable pattern recognition and generalizes better to unseen inputs. This preprocessing step is a foundational data engineering practice that directly determines how much signal versus noise the model learns from.

Data encoding converts data into a specific format or representation for compatibility purposes but does not remove errors, duplicates, or inconsistencies from training datasets.