Coupled prediction of fishery output value based on coupling coordination features and machine learning

In response to the neglect of the complex coupling relationship between fishing and aquaculture subsystems and the difficulty of quantifying interactive effects in traditional fishery output value prediction, which leads to insufficient prediction accuracy, this study is based on time-series data from 1986 to 2023. The entropy method is used to calculate the comprehensive development index for fishing and aquaculture, and then the coupling and coordination degrees are determined. The two are used as enhanced features, along with original variables (yield, operation intensity, aquaculture area, etc.), as inputs to the random forest model. Firstly, standardize the raw data and handle missing values; Secondly, construct a random forest model with coupled features and use Bayesian optimization for hyperparameter tuning; Simultaneously, set up a benchmark model without feature coupling for comparison. The results show that the optimized coupled enhanced random forest model has a coefficient of determination R^2=0.923 and a root mean square error of 692 million yuan, which is better than the benchmark model with R^2=0.891 and RMSE=874 million yuan. From 1986 to 2023, the coordination degree increased from 0.262 to 0.540, and the subsystems developed from severe imbalance to primary coordination. The feature importance analysis based on a random forest shows that the combined weight of the coupling degree and the coordination degree is significantly higher than that of any single yield indicator and is a key driver of fluctuations in the output value. The use of artificial intelligence methods to quantify subsystem interactions provides data support for short-term fishery management decisions and medium- to long-term industrial planning and enhances prediction reliability by identifying coupling features.