Benchmarking imputation methods for categorical biological data

Trait datasets are at the basis of a large share of ecology and evolutionary research, being used to infer ancestral morphologies, quantify species extinction risks, or evaluate the functional diversity of biological communities. These datasets, however, are often plagued by missing data, for instance, due to incomplete sampling, limited data and resource availability. Several imputation methods exist to predict missing values and recent studies have explored their performance for continuous traits in biological datasets. However, less is known about the accuracy of these methods for categorical traits. Here we explore the performance of different imputation methods on categorical biological traits combining phylogenetic comparative methods, machine learning and deep learning models. To this end, we develop an open‐source R package, to impute trait data while integrating a simulation framework to evaluate their performance on synthetic datasets. We run a range of simulations under different missing rates, mechanisms, biases and evolutionary models. We propose an integration between phylogenetic comparative methods and machine learning imputation, and an ensemble approach, in which selected imputation methods are combined. Our simulations show that this approach provides the most robust and accurate predictions. We applied our imputation pipeline to an incomplete trait dataset of 1015 elasmobranch species (i.e. sharks, rays and skates) and found a high imputation accuracy of the predictions based on an expert‐based assessment of the missing traits. Overall, our R package facilitates the comparison of multiple imputation methods and allows robust predictions of missing trait values. Our study highlights the benefits of coupling phylogenetic evolutionary models with machine learning inference to augment incomplete biological datasets.