Comparative Analysis of Statistical and Machine Learning Approaches for Predicting Fish Length from Otoliths

Christian James Cabingabang Morales^*Sanny David LumaynoRicardo Pamittan Babaran

• College of Fisheries and Ocean Sciences, University of the Philippines Visayas, Miagao, Philippines

This study assessed the predictive capacity of otolith morphometric variables for estimating fish length across six pelagic and demersal species from major Philippine fishing grounds. Using linear and nonlinear regressions, generalized additive models (GAMs), and machine learning (ML) algorithms, we evaluated 11 otolith morphometric and shape metrics to quantify species-level and ecological drivers of otolith–somatic scaling. Otolith length (OL) and otolith area (OA) consistently provided the strongest predictive power, whereas otolith perimeter (OP) performed weakest. Demersal species and the midwater schooling Decapterus kurroides showed highly predictable otolith–length relationships (𝑅² > 0.95), reflecting stable habitats and relatively uniform growth dynamics. In contrast, Selar crumenophthalmus and Thunnus albacares exhibited weaker predictability (R² ≤ 0.70), influenced by dynamic thermal regimes, variable prey fields, and ontogenetic shifts that introduce plasticity in otolith accretion. ML models, particularly Random Forest, outperformed classical approaches for species with heterogeneous growth patterns, capturing nonlinearities and interactions among morphometric variables. These results demonstrate that OL and OA serve as robust, transferable predictors for otolith-based size estimation in tropical multispecies fisheries. Integrating ecological context with flexible modeling frameworks can improve accuracy, especially for pelagic species inhabiting dynamic environments. The findings support the application of otolith morphometrics to stock assessment, bycatch monitoring, and automated length reconstruction workflows, and highlight the need for expanded sampling and advanced ML tools in future work.