OneHotEncoding

OneHotEncoding is a data encoding technique used to convert categorical variables into a numeric binary format. For a feature with k distinct categories, one-hot encoding creates k binary features. Each observation has a single 1 in the column corresponding to its category and 0s in all other columns. This representation avoids implying any ordinal relationship between categories.

Example: a color feature with categories red, green, and blue becomes three columns: color_red, color_green, color_blue. A red observation is (1, 0, 0), green is (0, 1, 0), and blue is (0, 0, 1). The resulting data is typically stored as a sparse matrix to save memory when many categories are present.

Applications and implementation: One-hot encoding is common preprocessing for machine learning models that require numeric input, including linear models and neural networks. It is supported by many libraries, such as pandas (get_dummies) and scikit-learn (OneHotEncoder). When handling binary categories, some workflows drop one column (dummy coding) to avoid redundancy and multicollinearity. Some implementations offer options to handle unknown categories encountered during inference.

Advantages and limitations: The method preserves nominal meaning without introducing artificial order. However, it increases dimensionality, producing sparse matrices that can be memory-inefficient for high-cardinality features. It also requires a consistent encoding scheme between training and deployment. In cases with many categories, alternatives such as target encoding, embedding, or hashing tricks may be preferred.