AI-powered decision support system for mariculture: Real-time fish mortality prediction with random forest

Fish mortality is a significant issue in mariculture, affecting productivity and sustainability. Predicting mortality risk in real-time is crucial for improving decision making and operational efficiency in mariculture management. This paper presents the development of a real-time fish mortality risk prediction model, designed as part of a Decision Support System using the Random Forest machine learning algorithm. The innovative aspect of this study lies in the real-time processing of sensor data to deliver daily mortality risk predictions, allowing for immediate adjustments to management practices. This study integrates water quality parameters (seawater temperature, salinity, conductivity, chlorophyll-a, turbidity, and dissolved oxygen) monitored through a sensor network, with daily fish mortality records input by farmers. The Random Forest model predicted fish mortality risk across five levels with an overall accuracy of 78.6 % and precision exceeding 70 % for each level. The model's feature importance analysis highlights seawater temperature, salinity, and turbidity as key predictors of fish mortality risk. This system supports fish farmers and site managers in daily operational decision making, particularly regarding feed and labor management. Future improvements in data collection and continuous model updates are expected to enhance the accuracy and utility of the Decision Support System in mariculture management. The graphical abstract illustrates the real-time fish mortality prediction system for mariculture. It depicts the integration of sensor networks collecting water quality data, a Random Forest algorithm analyzing these inputs, and a Decision Support System providing mortality risk levels. This system enables fish farmers to optimize feed and labor management, enhancing productivity and sustainability in mariculture operations. • Real-time fish mortality risk model for mariculture developed using Random Forest. • Sensor network tracked water quality and mortality to improve decision making in fish farming. • Model reached 78.6 % accuracy with precision above 70 % across five risk levels. • Seawater temperature, salinity, and turbidity were identified as the key predictors of fish mortality in the target area. • The system supports daily operational decisions for fish farmers and site managers, such as feed and labor management.