Implementation of Artificial Neural Network for IoT-Based Water Quality Classification in Fish Ponds

Abstract

This study presents the implementation of an Artificial Neural Network (ANN) to classify water quality in fish ponds using a dataset derived from a fuzzy inference-based IoT system. The previous fuzzy system utilized three sensor parameters—pH, Total Dissolved Solids (TDS), and temperature—to determine water quality (good, moderate, poor) through rule-based reasoning. Although the fuzzy approach produced accurate and interpretable results, it lacked adaptability to new data variations and required manual rule adjustments. In this research, the ANN model was trained using MATLAB’s Neural Network Toolbox with 120 dataset samples obtained from the fuzzy system’s outputs. The model architecture consisted of three input neurons (pH, TDS, temperature), one hidden layer with ten neurons using a tansig activation function, and one output neuron with purelin. Training of the model was conducted using the Levenberg–Marquardt backpropagation algorithm, employing a dataset split of 80% for training, 10% for validation, and 10% for testing. The results showed that the ANN achieved a classification accuracy of 94.8%, with a Mean Squared Error (MSE) of 0.85942 and a regression coefficient (R) of 0.94, indicating a strong correlation between predicted and actual data. Compared to the fuzzy inference method, the ANN model demonstrated better adaptability to unseen data and a higher level of generalization. This system can be integrated into IoT-based monitoring platforms for real-time, intelligent, and adaptive water quality prediction to support sustainable aquaculture.