Data must be interpreted correctly to understand the relevant information it carries. Especially today, data has massive content that needs to be retrieved by a specialist to be evaluated and validated before deciding on a solution. These massive data are commonly executed by computers trained through various specialized algorithms. Feature selection (FS) is critical in modern machine learning frameworks. Metaheuristic techniques can efficiently carry out FS to reduce the data dimension. In this paper, feature selection is carried out by using a K-Nearest Neighbors (KNN) wrapper with bird-based metaphor algorithms. Five different bird-based optimizers, namely, Cuckoo Search, Harris Hawks, Crow Search, Stain Bowerbird and Emperor Penguin, are considered for the study. For analyzing the different algorithms, six different types of datasets are used. The performance of an average number of features selected (AFS), accuracy, fitness, convergence capabilities and computational cost is compared.