The study compared nine machine-learning algorithms across 32 Philippine region-sector panels from 2002–2025 and used SHAP to explain environmental predictor contributions.
Key findings
- Kernel or neural architectures won 26 of 32 panels and scored 12.7% above tree ensembles on average. Partial pressure of CO2 ranked highest, while model and feature importance varied regionally.
Why this matters globally
Fisheries worldwide face environmental volatility. Region-specific and explainable models may aid planning, but SHAP explains model behaviour rather than ecological causality.
Thai researcher contribution
Walailak University-affiliated authors contributed the forecasting and explainable-AI framework for Philippine fisheries.
Limitations to consider
Forecast errors remain substantial, trend extrapolation is vulnerable to regime change, and fishing effort, policy or economic drivers may be incomplete. CO2 importance is not causal proof.
Verify the original sources
Machine Learning and Knowledge ExtractionRead the original article↗DOI: 10.3390/make8070197