K-means Clustering for Customer Segmentation in BigQuery ML

What K-means clustering is:

K-means clustering is an unsupervised machine learning algorithm built directly into BigQuery ML. It partitions a dataset into a specified number ($k$) of distinct, non-overlapping clusters or segments based on the mathematical similarity and distance between feature values of individual data points.

How it meets the requirements:

• Unsupervised segmentation (✓): The scenario explicitly states that the customer dataset does not have predefined target categories or historical labels. K-means is specifically designed for unlabeled data exploration and pattern discovery.
• Customer profiling (✓): K-means identifies natural groupings based on multidimensional feature vectors, such as visit frequency, annual spending, and order value.
• SQL native execution (✓): In BigQuery ML, you define MODEL_TYPE='KMEANS' inside the OPTIONS clause of a CREATE MODEL SQL statement without needing separate data export or external training infrastructure.

Key advantages and capabilities:

• No labeled data required: Training does not require training/evaluation data splits or predefined target labels (input_label_cols).
• Automatic feature preprocessing: BigQuery ML automatically standardizes numerical features and handles one-hot encoding for categorical variables during model training.
• Built-in cluster inspection: Once trained, you can inspect cluster centroids and analyze customer segment attributes using functions like ML.CLUSTERS and ML.PREDICT.
• Scalable in-place training: Operates directly over large BigQuery tables using standard SQL queries and managed cloud compute resources.

Why this is optimal:

K-means clustering is the industry-standard algorithm within BigQuery ML for data segmentation and customer grouping tasks where ground-truth outcome labels are unavailable.