Comparison of LSTM, RNN, and GRU Methods in Predicting Water Quality in the Totality Integrated Smart Aquaculture System

Shrimp farming in aquaculture holds significant global market potential. However, shrimp export growth in Indonesia, especially in West Nusa Tenggara (NTB) Province, declined in 2021 due to suboptimal aquaculture management and poor pond water quality. Pond water quality, involving parameters like temperature, pH, salinity, and dissolved oxygen (DO), is critical for successful shrimp farming. Therefore, an integrated automated system using the Internet of Things (IoT) is proposed to regularly collect and monitor water quality data. This study compares the effectiveness of Long Short-Term Memory (LSTM), Recurrent Neural Networks (RNN), and Gated Recurrent Unit (GRU) methods in predicting pond water quality parameters using data from 2021. The data is gathered through IoT-based devices and stored in a database, accessible via a web page with real-time graphs. The performance metrics of LSTM, RNN, and GRU models show that the GRU model outperforms the others across all evaluated criteria, including Mean Squared Error (MSE), Mean Absolute Error (MAE), Mean Absolute Percentage Error (MAPE), and Validation Accuracy (VAL). GRU achieves the lowest MSE (0.056058), MAE (0.149909), and MAPE (14.99%), as well as the highest validation accuracy (85.01%). This indicates that GRU provides the most accurate and reliable predictions. Integrating these prediction models into a web platform enables quick identification of issues and timely interventions, which can enhance the health and productivity of shrimp ponds. This project is expected to improve the efficiency and sustainability of the shrimp aquaculture sector in Indonesia by leveraging advanced predictive modeling techniques.