Which of the following statements is true about using binary values in k-Means?

• A. Binary values cannot be used in k-Means
• B. Binary values are required in k-Means
• C. Binary values must be scaled before use
• D. Binary values can be used in k-Means

Answer: (D)

Binary values can be used in k-Means because the method operates on numeric features and minimizes the sum of squared distances to cluster centroids. When you have 0/1 features, the centroid of a cluster becomes the mean of those values for the members in that cluster, giving a vector with components between 0 and 1. This fractional centroid can still be used to compute distances to each data point, and each point is assigned to the nearest centroid. Interpreting the centroid components as the probability that a feature is 1 in that cluster helps you understand the result. Since 0 and 1 are already on a standard scale, scaling binary values isn’t required, though you might scale if you’re mixing binary features with other types of features. Hence, binary values can be used in k-Means.