Abstract: In order to solve the problem of determining the appropriate feeding amount in Cynoglossus semilaevis aquaculture, researchers made the feeding rate table by collecting a lot of experimental data. However, the feeding rate table has problems, such as insufficient precision, reliance on manual experience and so on, which is not conducive to the realization of standardization and automation in feeding processes. To overcome this problem, this study collected the actual breeding record data of breeding enterprises and used machine learning methods to build a prediction model for the feeding amount of Cynoglossus semilaevis. The model was characterized by 29 items of information related to the feeding of Semilaevis. Firstly, Pearson correlation coefficient, Spearman correlation coefficient and Kendall correlation coefficient were used for correlation analysis. Secondly, principal component analysis method was used to reduce the data dimension. Then the dimensionally reduced data set was randomly divided into training set, verification set and test set in a ratio of 7∶2∶1, which were used to train the model, adjust model parameters and evaluate model performance respectively. In order to explore the prediction effect of machine learning models based on different algorithms on the amount of bait, BP neural network, support vector machine and random forest models were constructed on the training set. The average absolute error of the BP neural network model on the test set was（MAE）0.210, the root mean square error（RMSE）was 0.301, and the coefficient of determination（R²）was 0.802; The MAE of the support vector machine model on the test set was 0.248, the RMSE was 0.370, and the R² was 0.701; Of the random forest model on the test set the MAE was 0.106, the RMSE was 0.185, and the R² was 0.925. The results demonstrate that among the three models, the random forest model which performs best in feeding amount prediction, is the most suitable one for the feeding deciding task in actual Cynoglossus semilaevis aquaculture, and is conducive to the realization of standardization and automation in feeding processes.

Key words: machine learning, Cynoglossus semilaevis, feeding amount