A machine learning ensemble approach for predicting growth of abalone reared in land-based aquaculture in Hokkaido, Japan

Abstract

Land-based aquaculture is an ideal aquaculture solution for creating high-quality seafoods and providing optimal conditions for maximizing growth of seafood production because environmental factors are well controlled. Predicting the growth of indoor-cultured abalone is meaningful because it facilitates evaluation of the effectiveness of this type of farming and understanding of the effects of controllable environmental factors on abalone growth. In this study, such predictions were made using an ensemble of machine learning algorithms: the random forest, gradient boosting, support vector machine, and neural network algorithms. Data were collected in the town of Fukushima, Hokkaido, Japan, and the increase in the weight of abalone was hypothesized from independent variables, including air and water temperature, loss of individuals caused by mortality or emigration, flow speed, age, and growth period between two measurements. The results showed that the ensemble method predicts growth well, with a low mean absolute error and mean square error. Temperature adjustment can make a strong contribution to increasing the weight of abalone, where a stable and warm temperature enhances growth. Moreover, the age of abalone is closely related to growth. Abalone size increased strongly in the early stages but decreased slightly once near market size.