Prediction of Tl(I) adsorption onto metal oxides and identification of critical factors using a machine learning-based model.

Thallium is a highly toxic element, which is widely found all over the world. Adsorption is one of the most common techniques for thallium removal. Traditional adsorption studies face several limitations, such as a limited ability to predict adsorption results, inability to concurrently deal with the effects of multiple factors on adsorption outcomes, and fail to quantify the importance of these factors. Machine learning (ML), as a novel technology, can overcome those limitations and make remarkable achievements in this field. However, to the best of our knowledge, no research had been conducted on the application of machine learning to remove Tl. In this study, the ML prediction models were developed using 414 data points collected from more than 30 articles about the adsorption of Tl by metal oxide materials that can be searched at present. Meanwhile, this prediction model simultaneously studied the environmental conditions (such as solution pH, reaction temperature, reaction time, initial Tl concentration), adsorbent dosage on the adsorption results and the methods of out-of-bag error and Pearson correlation coefficient were used to quantify the importance of these factors. It found that pH and C contributed more than 50 % on Tl adsorption. After a comprehensive comparison among three chosen models, i.e., Random Forest (RF), Least Squares Support Vector Machine (LSSVM) and Particle Swarm Optimization-Support Vector Machine (PSO-SVM), LSSVM showed excellent prediction performance (R >0.9). Overall, this study innovatively introduced machine learning into the study of Tl adsorption by metal oxide and obtained a universal model, which provided a powerful tool for subsequent prediction of adsorption efficiency. It is crucial to acknowledge that, owing to limitations in data collection, the material properties were excluded from this study. Future research is expected to be more comprehensive, incorporating the influence of material properties on adsorption capacity to enhance the model's universality.