A Deep Learning Approach to Predict Dissolved Oxygen in Aquaculture

Abstract

Fish is one of the major sources of protein nutrients for people. Most fish supply comes from the natural habitants which include rivers, lakes, seas and oceans. However, the high demand has necessitated fish farming from man-made lakes, ponds and swamps. There are various issues that pose risks to fish survival and growth, and among these include the level of dissolved oxygen (DO) in the water which is an essential environmental condition whose scarcity leads to suffocation of fish and ultimately death. This study aimed at designing a prediction model for DO in aquatic environments. To achieve the objective, time series data consisting of 70374 records and 15 attributes from Mumford Cove in Connecticut, USA collected for over 5 years was preprocessed and used to train long-short term memory (LSTM) recurrent neural network (RNN) for DO prediction. The training and testing data were obtained by splitting the dataset into 70% and 30%, respectively. Regression models include linear regression (LR), support vector regression (SVR) and decision tree regression (DTR) were also created for comparisons. The performance of the models was evaluated on the basis of mean absolute percentage error (MAPE), mean squared error (MSE), mean absolute error (MAE) and coefficient of determination ($\mathbf{R^{2}}$ score). LSTM achieved superior performance compared to the regression models. Conclusively, DO on such multivariate time series data can be well achieved with LSTM RNN.