SAUDI JOURNAL OF BIOLOGICAL SCIENCES, vol.30, no.4, pp.1-9, 2023 (SCI-Expanded)
Diseases are quite common in fish farms because of changes in
physico-chemical characteristics in the aquatic environment, and
operational concerns, i.e., overstocking and feeding issues. In the
present study, potential factors (water physico-chemical characteristics
and heavy metal load) on the disease-causing state of the pathogenic
bacteria Lactococcus garvieae and Vagococcus sp. were
examined with machine learning techniques in a trout farm. Recording of
physico-chemical characteristics of the water, fish sampling and
bacteria identification were carried out at bimonthly intervals. A
dataset was generated from the physico-chemical characteristics of the
water and the occurrence of bacteria in the trout samples. The eXtreme
Gradient Boosting (XGBoost) algorithm was used to determine the most
important independent variables within the generated dataset. The most
important seven features affecting bacteria occurrence were determined.
The model creation process continued with these seven features. Three
well-known machine learning techniques (Support Vector Machine, Logistic
Regression and Naïve Bayes) were used to model the dataset.
Consequently, all the three models have produced comparable results, and
Support Vector Machine (93.3% accuracy) had the highest accuracy.
Monitoring changes in the aquaculture environment and detecting
situations causing significant losses through machine learning
techniques have a great potential to support sustainable production.