An Up-to-date Approach Using Machine Learning Methods in Fish Condition Factor Estimation

The present study aims to estimate the condition factor (CF) of mackerel (Trachurus mediterraneus, Steindachner, 1868) by making use of three input parameters (length, weight, and sex) that the CF is related to. For this purpose, data were obtained from 866 mackerel fished in the Eastern Black Sea. In the present study, the estimation performances of Multiple Linear Regression (MLR), Levenberg-Marquardt (LM), and Gaussian Process Regression (GPR) models, among statistical instruments, were compared. Quality levels of the models were compared by making use of the coefficient of determination (R2), the root mean square of error (RMSE), and the mean absolute percentage error (MAPE) criteria. It was aimed to select the model, which yields the best estimation performance for length-weight relationships, by comparing the verification results. The results showed that the ANN trained with the GPR model yielded the highest accuracy. It was determined that the R2 of estimation results achieved using the GPR model was higher than 0.99 for all the parameters. Given these results, it can be concluded that the GPR model that is suggested here is a robust instrument to estimate CF at a high level of accuracy.