Enhancing Length at First Maturity Estimation Using Machine Learning for Fisheries Resource Management: A Case Study on Small Yellow Croaker (<i>Larimichthys polyactis</i>) in South Korea

Abstract

Small yellow croaker (Larimichthys polyactis) is a critical economic fish species in South Korea, where effective management is essential due to concerns over declining populations. This study aims to enhance fishery management strategies by applying machine learning techniques to classify the maturity stages and estimate the length at first maturity (L50 and L95), comparing these results with those obtained using traditional macroscopic methods. Five machine learning models, including Decision Tree (DT), Random Forest (RF), LightGBM (LGBM), EXtreme Gradient Boosting (XGB) and Support Vector Machine (SVM), were developed and evaluated for their effectiveness in predicting maturity stages. The XGB model demonstrated superior performance with the highest evaluation final score and low computation time. Using generalized linear models (GLM), this study estimated L50 and L95 for both machine learning predictions and macroscopic observations. The results showed that machine learning models, particularly XGB, provided more precise estimates with narrower confidence intervals and better model fit than the traditional macroscopic methods. These findings can support more sustainable fisheries management practices by offering reliable tools for setting appropriate regulatory measures, such as minimum landing sizes, which contribute to the conservation of marine resources.

Keywords

• length at first maturity
• machine learning
• macroscopic methods
• fisheries management
• small yellow croaker