Integration of RNN and CatBoost models in a tea-waste biochar filtration system for toxic organic pollutant removal efficiency prediction

Water pollution is a dreadful global crisis undermining the environment and economy. In order to combat this issue, several methods and techniques are adopted for treating the polluted water. Adsorption by biowastes is one of the most economically viable, simple, and effective methods for wastewater treatment. In spite of numerous reports in the literature showing the removal of various pollutants, there is still room for investigation in the field of simultaneous adsorption of varied categories of pollutants. The present study focuses on the simultaneous removal of organic water contaminants like dyes, agrochemicals, and aromatic compounds from wastewater using biochar prepared from tea waste as adsorbent. A detailed investigation on the effect of contact time, pH, dosage, and temperature on the adsorption performance of adsorbent has been carried out. At optimized reaction condition of 5 mg ml^-1 of adsorbent dosage at pH 2 for 60 min, 82.66% overall removal was obtained for 40 ppm of pollutant (malachite green, congo red, chlorpyrifos, and 4-nitroaniline) mixture. Further, the percentage removal was predicted using two machine learning (ML) models: CatBoost and Recurrent Neural Network (RNN), with and without Bayesian optimization. The prediction capability of these models was evaluated using three performance metrics: coefficient of determination (<i>R</i> ²), Mean Absolute Error (MAE), and Root Mean Square Error (RMSE). Based on the evaluation, RNN was found to be the most effective model for % removal prediction based on higher <i>R</i> ² value of 0.960. Moreover, the fabrication of a portable column filtration device for the removal of coexisting harmful organic pollutants has been demonstrated. The results confirm that tea waste (TW)-derived biochar, coupled with advanced machine learning models, is a promising solution for real-time wastewater treatment.