World Aquaculture Singapore 2026

June 2-5, 2026

Singapore

LEVERAGING GENERATIVE ARTIFICIAL INTELLIGENCE FOR ACCESSIBLE PRECISION AQUACULTURE: AN INTERACTIVE INTERFACE FOR SHRIMP SURVIVAL PREDICTION

Neil Arvin Bretaña*

IU International University of Applied Sciences

Mülheimer Str. 38, 53604 Bad Honnef, German y

Neil-arvin.bretana@iu.org



The rapid expansion of aquaculture necessitates innovative solutions to manage resources and maximize productivity. While machine learning models, such as Support Vector Machines (SVM), have demonstrated high accuracy in predicting Penaeus vannamei survival rates, their practical adoption is often hindered by the technical complexity of interpreting model outputs. This study builds upon prior modeling of industry data from nine shrimp ponds in the Philippines to develop a Generative Business Intelligence (GenBI) interface.

Our previous research established that an SVM model could predict survival rate classes (threshold set at 95% for balanced distribution) with 84.69% accuracy using a comprehensive dataset of physio-chemical properties, feed data, and water management inputs. We integrated a Large Language Model (LLM) framework to act as a natural language intermediary. This "GenBI" layer allows users to interrogate the underlying SVM and Random Forest models using natural language queries. The dataset utilized for this integration includes 3 harvest cycles, capturing variables such as Dissolved Oxygen (DO), temperature, and specific supplements like Top S (immunostimulants) and Pro 2 (gut wellness).

The system allows users to ask questions like "Which water quality parameters are currently limiting survival in Pond 4?" The LLM retrieves the most significant variables - often physio-chemical markers which previously showed high true positive rates - and translates them into actionable advice. By bridging the gap between "black-box" data and farm management, this tool promotes sustainable practices through immediate, evidence-based adjustments to pond conditions.

Integrating real-time water quality monitoring with generative algorithms unlocks potential for dynamic yield predictions, further revolutionizing sustainable and efficient practices in the rapidly growing aquaculture industry.