Accuracy assessment of RFerns, NB, SVM, and kNN machine learning classifiers in aquaculture

Abstract

BackgroundFish diseases are the primary cause of decreased productivity and financial losses in fish farms. Detecting and monitoring fish disease using human eye is a time-consuming manual process. Technological advancements have made possible that data may now be collected at unprecedented speeds, quantities, and complexity, with far less effort and expense. Species must be provided optimum environmental conditions for healthy production. However, sub-optimal conditions and managerial issues could lead to disease outbreaks in fish farms. Machine learning (ML) classifiers can provide solutions to fish farm difficulties by collecting data with less efforts. MethodsThis study investigated water physico-chemical parameters potentially responsible for bacterial disease outbreak in fish farms. Four most popular ML algorithms, i.e., support vector machine (SVM), naïve bayes (NB), random ferns (RFerns), and K-nearest neighbor (kNN) were used to detect the physico-chemical parameters of water causing the disease. Data were collected from 3 different farms in two-month periods for 1 year. Models were developed by using 10-fold cross validation procedure to the training dataset data for each model. The models were examined using seven distinct metrics throughout the training and testing phases. ResultsThe SVM and RFerns classifiers produced accurate results (100% for both) during the testing phase, while kNN and NB classifiers achieved lower accuracy (91.3% accuracy for both). ConclusionThe SVM and RFerns algorithms performed better than kNN and NB algorithms in both the training and testing phases of the study. Although earlier research confirms the efficacy of the SVM algorithm in aquaculture, comparable efficacy of RFerns with SVM has been reported in this sector for the first time, which is a significant addition to the literature.