Enhancing Neural Network Performance for Water Quality Forecasting with Principal Component Analysis in Intensive Aquaculture

Abstract

This work investigates the application of Principal Component Analysis (PCA) to enhance the performance of a neural network regression model for water quality forecasting in intensive aquaculture. The dataset from an intensive cultivation study includes daily readings of controlled pond biochemical parameters. The standard and PCA-enhanced models had their performance evaluated based on the MSE, MAE, and R². The results demonstrate that the model incorporating PCA outperformed the standard model. The PCA model achieved lower training and testing MSEs, with a notable reduction in MAE. These findings highlight the effectiveness of PCA in improving the accuracy and efficiency of neural network models by reducing dimensionality and emphasizing the most informative features.