Predicting Fishing Grounds With Dimensionality Reduction and SHAP Analysis: The Case of Jumbo Flying Squid ( Dosidicus gigas ) Off Peru

ABSTRACT The jumbo squid ( Dosidicus gigas ), a pelagic cephalopod of significant economic and ecological importance off the coast of Peru, is highly sensitive to marine environmental shifts. This study synthesized jumbo squid fishery logbook data with vertical variables (0–147 m depth) and two large‐scale climate indices, the Pacific Decadal Oscillation (PDO) and the Niño 1 + 2 sea surface temperature anomaly, from 2014 to 2021. We assessed K ‐means clustering for vertical oceanographic data dimensionality reduction and trained and evaluated five tree‐based regression models using the SHAP (SHapley Additive exPlanations) framework for Dosidicus gigas distribution prediction and model interpretability. Our results demonstrated that K ‐means clustering identified three representative depth groupings from the environmental profiles, reducing the depth‐dependent features from 140 to 15 and generally maintained or improved predictive performance while substantially reducing computational time, with random forest (RF) and gradient boosting regression tree (GBRT) providing the strongest overall accuracy. Subsequent SHAP analysis identified near‐surface dissolved oxygen, the Niño 1 + 2 index, and the PDO index as the most influential predictors of Dosidicus gigas distribution during the study period. By reducing feature redundancy while maintaining model performance and improving computational efficiency, the framework provides a practical and interpretable strategy for simplifying complex vertical environmental data while facilitating the characterization of environmental associations underlying fishery habitat.