Finite‐Element‐Informed Pyramid Neural Network With Draco Lizard Optimizer for Accurate and Efficient Water Contamination Classification

ABSTRACT

Water Contamination is a major issue of concern in the aquaculture industry, and it affects the sustainability of the ecosystem as well as the health of the aquatic species. Measuring pollution clean‐up in water is the key component of efficient supervision and defense of aquatic organisms. Conventional ways of evaluating results are usually slow, expensive, and subject to errors, and the current methods of deep learning have certain issues to do with the variability of the data within the system being evaluated, interpretability, and predictive analysis. To mitigate the problems, this paper presents a Finite‐Element‐Informed Pyramid Neural Network with Draco Lizard Optimizer (F‐E‐IPNNet‐DLO) as a robust water contamination classifier. The benchmark dataset is preprocessed with Trimmed Scores Regression to K‐Means Clustering (TSRK‐MC) to deal with noise, outliers, and missing values, and then the Billiards‐inspired Ebola Search Optimization Algorithm (B‐ESOA) to predict features. The chosen features are then transferred to a pyramid‐featured neural network to extract the features, multi‐scaled using Draco Lizard Optimizer (DLO). Exceptional accuracy, recall, precision, F1‐score, and specificity (99.97% on Dataset‐1 and 99.96% on Dataset‐2) and minimal predictive errors demonstrate the reliability, applicability, and efficiency of the model in sustainable aquaculture and environmental conservation.