A Decision Support System for aquaponics prediction based on the Intelligent Internet of things

Abstract

Aquaponics is an emerging farming technique. Managing and optimising aquaponics systems is complex and requires expertise in aquaculture, hydroponics, and microbiology. Effective decision-making is crucial to maintaining optimal conditions for plants and fish so the system can thrive. Current research emphasises water quality monitoring but lacks the analysis of key parameters and their impact on plant growth and system productivity. There is a need for data driven solutions to help users, especially beginners, optimise resource use and enhance performance. The research aimed to develop a decision support system (DSS) for aquaponics that provides data-driven insights into plant growth and water quality using Explainable Artificial Intelligence (XAI). The following research objectives were used to achieve this: 1) Identify key parameters for monitoring plant growth and water quality. 2) Develop machine learning (ML) prediction models. 3) Evaluate the performance of different ML algorithms using regression metrics. 4) Design and develop a machine learning-based decision support system to facilitate decision making in aquaponics. 5) Assess the decision support system’s usability from the aquaponics stakeholders’ perspective. This study adopted an objectivist ontological stance to determine the feasibility of developing a DSS for aquaponics prediction. The epistemological stance was positivism. To meet the objectives, a deductive research approach was adopted with a quantitative methodological choice. The data parameters collected are plant height, plant diameter, Potential of Hydrogen (pH), Total Dissolved Solids (TDS), water temperature, ambient temperature and humidity. An experimental design was used to train and evaluate several supervised ML algorithms: linear regression, random forest, K-Nearest Neighbor (KNN), eXtreme Gradient Boosting (XGBoost), and Multi-Layer Perceptron (MLP). These models were assessed using the regression metrics Mean Squared Error (MSE), Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), R-squared (R²), and Adjusted R-squared. The results revealed that both random forest and XGBoost achieved the best performance for plant diameter prediction with MSE = 0.00, RMSE = 0.05, and MAE = 0.03 with R² and Adjusted R² scores of 94%. In plant height prediction, random forest performed well with MSE = 0.00, RMSE = 0.06, and MAE = 0.05, along with a high R² of 93% and Adjusted R² of 92%. XGBoost performed well in pH prediction with MSE = 0.02, RMSE = 0.13, and MAE = 0.09, along with high R² and Adjusted R² of 79%. In TDS prediction, linear regression performed well with MSE = 0.00, RMSE = 0.01, and MAE = 0.01, along with perfect R² and Adjusted R² scores of 100%. A DSS was developed using the FLASK framework to predict plant height and diameter, water pH, and TDS. SHapley Additive exPlanations (SHAP) was used to enhance transparency by showing each feature's impact on predictions. The usability of DSS was evaluated by aquaponics stakeholders through the System Usability Scale (SUS) by. The DSS obtained a usability rating of 72%, which indicates an acceptable level of usability. Theoretically, the study demonstrates applying ML and XAI to predict plant growth and water quality under South African conditions. Methodologically, it offers a structured approach to integrating ML, Internet of Things and AI in aquaponics. Practically, it delivers a DSS to help practitioners monitor and optimise key parameters, improving overall system performance and outcomes.