KERNEL PCA AND ENSEMBLE LEARNING FOR PREDICTING WATER CHEMISTRY AND MICROBIOLOGY PROPERTIES OF PONDS IN PENAEUS VANNAMEI CULTIVATION

Abstract

Water quality is one of the important factors that determine shrimp cultivation yields. It determines shrimp growth and survival rate. Hence water quality monitoring is one of the important activities in shrimp farming. Despite its importance, monitoring water quality during shrimp farming can be costly. This research was conducted to develop prediction models that would give farmers insight about water quality of their ponds. The prediction models used temperature, dissolved oxygen, salinity, and pH as input to predict chemical and microbiological properties of the water. The chemical properties included hardness, magnesium, calcium, and total ammonia whereas the microbiological properties included total organic matter and total plankton. The prediction model was built by combining Kernel Principal Component Analysis and machine learning algorithms (Random Forest and Gradient Boosting separately). The method was tested on the data collected from 31 ponds. The results showed that the algorithm can predict biological and chemical conditions of water (Total Organic Matter, Hardness, Calcium, Magnesium) quite well with R2 score higher than 0.8 on most parameters. Further the result also showed that the combination of Kernel PCA (configured with 2 order polynomial kernel) and Gradient Boosting had best prediction accuracy. These findings show that the method can be used as an alternative to laboratory tests. This would help the farmer in monitoring their pond’s condition in a faster and less expensive way. This also would help farmers who don’t have access to laboratory facilities in monitoring the water quality condition.