Integrating biological and machine learning models for rainbow trout growth: Balancing accuracy and interpretability

Area of Science:

• Ecology
• Computational Biology
• Fisheries Science

Background:

• Predictive modeling for invasive species management requires balancing accuracy and ecological interpretability.
• Traditional models often struggle to capture complex environmental interactions influencing species growth.

Purpose of the Study:

• To evaluate hybrid frameworks integrating biological and machine learning models for predicting rainbow trout growth.
• To compare the performance of traditional, Bayesian, and machine learning models using tag-recapture data and environmental covariates.

Main Methods:

• Utilized ten years of tag-recapture data for rainbow trout (Oncorhynchus mykiss) in the Lower Colorado River.
• Compared traditional and Bayesian von Bertalanffy (VBGM) and Gompertz models with Random Forests, XGBoost, LightGBM, Support Vector Regression, Neural Networks, and ensemble methods.
• Employed probabilistic performance analysis and feature importance analysis.

Main Results:

• Hybrid models achieved 70-80 percent error reductions compared to baseline models.
• A stacked ensemble of XGBoost and VBGM showed the best performance (RMSE = 15.96 mm).
• Gradient boosting models (LightGBM, XGBoost) and Bayesian Model Averaging also demonstrated high accuracy and uncertainty quantification.

Conclusions:

• Hybrid ensemble frameworks unite predictive performance with mechanistic insight for ecological forecasting.
• These models provide a generalizable template for systems requiring both accuracy and interpretability in ecological predictions.
• Incorporating environmental context and advanced modeling significantly enhances the accuracy of fish growth predictions.