Machine Learning Based Prediction of Catch Yields in Small Scale Fisheries

Abstract

Small-scale fisheries are critical for food security and livelihoods but increasingly face threats from overfishing and environmental change. This study employed machine learning to predict catch yields in Cogtong Bay, Philippines, using 3,478 records collected in 2020–2021 that included species, gear, location, season, and weather data. Models developed in Python with scikit-learn and XGBoost compared Random Forest (RF), Extreme Gradient Boosting, support vector machines, and logistic regression. Trips were classified as high- or low-yield based on total catch weight. Results indicate that RF achieved the highest performance (≈90% accuracy), correctly identifying more than 90% of high-yield trips and outperforming other models. Feature importance analysis revealed gear type and season as the strongest predictors, with location and weather exerting secondary influence. These findings demonstrate that ensemble machine learning models can capture complex fisheries dynamics and provide reliable decision support. The approach offers a scalable tool for adaptive, data-driven fisheries management and more sustainable, resilient coastal livelihoods.