Cooperative Predation in the Sardine Run: A Multi-Objective Bio-Inspired Algorithm for Robust Feature Selection in High-Dimensional Spectroscopy

High-dimensional spectroscopic data present a persistent challenge for chemometrics, demanding featureselection methods that are both predictive and parsimonious. Inspired by the multi-species predation of asardine run, we introduce HSHFSM (Hybrid Sardine Hunting Feature Selection Method), a novel multiobjective metaheuristic. HSHFSM simulates the distinct hunting strategies of dolphins (exploitation),seabirds (balanced search), and swordfish (disruption) to navigate complex data landscapes. To ensurefull reproducibility, the entire framework, which was benchmarked against ten standard methods viarigorous nested cross-validation, is delivered in an open-source Google Colab application.The validation experiments reveal HSHFSM's powerful adaptability. On a synthetic sparse linear model,a specialized 'Dolphin-only' ablation demonstrated the efficacy of pure exploitation, achieving nearperfect feature recovery with an R² of 0.957. In stark contrast, on the highly collinear p >> n GasolineNIR dataset, the robustness of the full hybrid model proved essential, delivering a state-of-the-art R² of0.968 while reducing dimensionality by 73.8%. This demonstrated ability of HSHFSM to adapt itsdominant strategy to the underlying data structure represents a significant advance in building highlypredictive and interpretable models from complex spectroscopic data.