Despite its rapid growth, shrimp farming faces many challenges, such as diseases, water quality and environment, and climate change. These problems cause shrimp farming to have different survival rates and production levels in different ponds. Due to all of the factors that affect it, it is typically characterized by boom and bust cycles. The latter are usually caused by production crashes resulting from disease outbreaks. With all of the factors that increase uncertainty in shrimp farming, it is important to develop reliable methods for predicting the survival rate (SR) of P. vannamei in different scenarios.
In Indonesia, SR prediction is commonly done based on feeding rate or feeding program. The prediction merely considers the ratio of the amount of feed thrown to ponds towards its theoretical feeding that is estimated based on the feeding rate. This condition makes the SR prediction accuracy depend on the feeding estimation accuracy and how disciplined the farmer is in following the feeding program. Further, this method is only able to estimate the current state of the cultivation without giving any forecast. Based on our research, this method is only able to predict SR with an R2 score of 0.65. This research tries to solve the problem by using machine learning that makes predictions based on more diverse variables such as feed consumption, stocking density, mortalities, feeding control, targeted length of cultivation, moon phase, and shrimp growth to project SR on a specific targeted age of cultivation.
The dataset used in this study was collected from several locations in Indonesia and contains 867 shrimp cultivation cycles from 146 farms. The dataset was cleaned, imputed, and feature-engineered to make the prediction. This research compared 3 algorithms, Random Forest (RF), Gradient Boosting (GB), and Elastic network (EN). The results showed that the random forest model outperformed the other models, achieving an average R2 score of 0.853 over 10 trials. Meanwhile R2 for GB and EN model area 0.67 and 0.24. Those results indicate that the RF model can be used as a reliable tool for predicting the survival rate of vannamei shrimp and can help farmers and aquaculture managers optimize their production and cultivation strategies.