A retail company stores several terabytes of structured transaction data in BigQuery tables. The data analytics team needs to train and evaluate a binary classification model to predict customer churn. The team wants to implement this solution using their existing standard SQL skills, without exporting or moving the data outside the data warehouse. Which solution should the team use?

Use BigQuery ML to build and evaluate the classification model directly in BigQuery using standard SQL queries.: ### **What BigQuery ML is:** **BigQuery ML** is a Google Cloud service that enables data analysts and practitioners to build, train, evaluate, and operationalize machine learning models directly inside **BigQuery** using standard **SQL queries**. It eliminates the necessity of exporting large volumes of data to separate machine learning environments. ### **How it meets all requirements:** - **In-place training on large datasets (✓):** Trains models directly over multi-terabyte datasets residing inside BigQuery tables, avoiding costly and complex data extraction or movement pipelines. - **Standard SQL interface (✓):** Allows analysts to use familiar SQL syntax (such as the `CREATE MODEL` statement and `ML.EVALUATE` function) rather than requiring Python, R, or dedicated ML frameworks. - **Binary classification support (✓):** Natively supports classification models like **logistic regression**, boosted trees, and deep neural networks to predict binary outcomes (such as customer churn). - **Built-in evaluation and inference (✓):** Provides built-in SQL functions like `ML.EVALUATE` and `ML.PREDICT` to assess model metrics and generate batch predictions directly in SQL workflows. ### **Key advantages:** - **Faster time-to-market:** Eliminates data migration delays and the overhead of managing separate infrastructure. - **Automatic data preprocessing:** Automatically performs common feature engineering tasks like one-hot encoding for categorical data and normalization for numerical features. - **Familiar skillset:** Enables SQL-proficient analysts to develop production-ready machine learning models without learning specialized programming languages. - **Tight security and governance:** Keeps data within BigQuery's access control and security boundaries throughout the entire machine learning lifecycle. ### **Why this is optimal:** BigQuery ML is designed precisely for scenarios where structured data resides in BigQuery and teams wish to leverage SQL for end-to-end model creation, evaluation, and prediction without managing external compute clusters or export pipelines.