Cardinality counts the unique categories represented by a categorical feature. It guides encoding choices: many categories can make one-hot encoding unwieldy, while ordinal encoding may impose an unjustified order. section 8.4
Employee job titles
An employee dataset contains 385 employee-position titles, compared with two genders and 37 departments. The job-title feature is therefore high-cardinality; spelling errors and alternate titles can increase its apparent cardinality further. section 8.4
High-cardinality string features often require similarity-based encoding, which groups categories with similar wording rather than matching only exact strings. SimilarityEncoder, MinHashEncoder, and GapEncoder provide alternative treatments. section 8.4
