Investigating Shape Variation Using Generalized Procrustes Analysis and Machine Learning

The biological investigation of a population’s shape diversity using digital images is typically reliant on geometrical morphometrics, which an approach based user-defined landmarks. In contrast to this traditional approach, the progress in deep learning has led numerous applications ranging from specimen identification object detection. Typically, these models tend become black boxes, limits usage recent for applications. However, explainable artificial intelligence tries overcome limitation. This study compares explanatory power unsupervised machine landmark-based approaches population structure investigation. We apply convolutional autoencoders as well Gaussian process latent variable two Nile tilapia datasets investigate consensus clustering. factors were extracted and compared generalized Procrustes analysis. Hypotheses Bayes factor are formulated test unambiguity unveiled by models. findings show that it possible obtain biologically meaningful results relying learning. Furthermore we unveil structures close true clusters. found 80% clusters autoencoder significantly different remaining Similarly, 60% model different. conclude outperform analysis, where 16% cluster was be applied still have limited explainability. recommend further in-depth investigations used model.