Classifying Horizontal Swimming Behavior of Atlantic Salmon Using Deming Regression

Abstract

As aquaculture production of fish is progressively increasing, we need to understand the relationships between their swimming behavior and external variables in order to improve their welfare. Fish in sea cages normally exhibit circular swimming patterns, however, an efficient model to automatically identify circling as a frequent and no circling as an abnormal behavior is lacking. In this study, we used acoustic telemetry data to classify the swimming behavior of Atlantic salmon into three classes: slow circling, fast circling, and no circling. This was achieved by developing an algorithm that uses a locally weighted form of Deming regression. We also developed an interactive visualization tool by representing every fish with a horizontal stacked bar colored according to swimming classes which enables the analysis of individual and population-level swimming behavior. Finally, the impact of several external variables such as natural light, temperature, dissolved oxygen, water levels, weather conditions, and wind speed on each swimming class is analyzed.