Predicting and Analyzing Water Quality using Machine Learning for Smart Aquaculture

Abstract

The greatest threat to the sustainability of inland fishery resources is environmental degradation. Aquatic pollution, destruction of fish habitats, water abstraction and impacts on aquatic biodiversity are all increasing. Industrial aquaculture has resulted in a far-reaching variety of environments that threatens native wild fish populations. The spread of deadly diseases and parasites is another challenge of aquaculture. These challenges can be overcome by applying proper water quality monitoring. This research suggests potential improvements after analyzing earlier studies regarding water quality prediction. In this research, previous works on water quality prediction have analyzed and proposed a novel smart aquaculture system using machine intelligence (SAS-MI) model. The novel SAS-MI model uses the deep convolutional neural network (D-CNN) and k-means clustering technique. The k-means clustering used in the SAS-MI clusters the unlabeled dataset used for training and testing. The D-CNN predicts the quality of the water for smart aquaculture by automatic feature engineering by the neurons. The proposed model performance was evaluated using the comparative study performed over the existing prediction models such as logistic regression, decision trees, XG boost classifiers, k-Nearest Neighbors, and SVM. The SAS-MI model with D-CNN and k-means clustering gives significant results for prediction accuracy, F1-score, and Mean Square Error (MSE).