Harnessing neural network model with optimization for enhanced ciprofloxacin antibiotic adsorption from contaminated water: A transparent and objective framework

The presence of antibiotics in water sources poses a significant threat to environmental sustainability and human health. Addressing this issue necessitates the development of precise predictive frameworks. This study focuses on leveraging machine learning (ML) methodologies to establish a predictive model for determining the maximum removal efficiency of ciprofloxacin (CIP) antibiotic from contaminated water. Two optimization strategies, namely Bayesian optimization and random search multilayer perceptron (MLP), were employed to fine-tune the hyperparameters of neural networks for optimizing adsorption efficacy. The Bayesian optimization model exhibits remarkable performance in antibiotic adsorption, achieving correlation coefficients (R²) of 0.9985 and 0.9856 on training and testing datasets, respectively. Similarly, the random search MLP yields commendable results with R² scores of 0.9958 and 0.9650 for training and testing, respectively. An iterative particle swarm optimization (PSO) methodology was used to determine optimal parameters and validate them through experimentation. The predictive model forecasts a substantial antibiotic adsorption or removal efficiency (>97 %) under optimized conditions, involving 9.6 mg/L of CIP, 400 mg/L of CuWO₄@TiO₂ adsorbent, a contact duration of 40 min at ambient temperature, and a neutral pH (7.0). The transparency of the model was assessed through feature importance analysis, revealing treatment time and adsorbent doses as the most influential features in predicting CIP adsorption. The integration of machine learning models with nano adsorbent presents a promising approach for mitigating antibiotic contamination in water sources.