manydist - Unbiased Distances for Mixed-Type Data
A comprehensive framework for calculating unbiased
distances in datasets containing mixed-type variables
(numerical and categorical). The package implements a general
formulation that ensures multivariate additivity and
commensurability, meaning that variables contribute equally to
the overall distance regardless of their type, scale, or
distribution. Supports multiple distance measures including
Gower's distance, Euclidean distance, Manhattan distance, and
various categorical variable distances such as simple matching,
Eskin, occurrence frequency, and association-based distances.
Provides tools for variable scaling (standard deviation, range,
robust range, and principal component scaling), and handles
both independent and association-based category
dissimilarities. Implements methods to correct for biases that
typically arise from different variable types, distributions,
and number of categories. Particularly useful for cluster
analysis, data visualization, and other distance-based methods
when working with mixed data. Methods based on van de Velden et
al. (2024) <doi:10.48550/arXiv.2411.00429> "Unbiased mixed
variables distance".
