A data-driven machine learning approach for interpretable prediction of desalination stability of carbon materials for capacitive deionization

Capacitive deionization (CDI) has emerged as a highly efficient, energy-saving, and environmentally friendly desalination technology, typically employing carbon materials as electrodes. However, ensuring the desalination stability of carbon materials remains a critical challenge for their practical application. Traditional experimental methods for fabricating carbon materials with enhanced desalination stability are often both laborious and time consuming. Machine learning (ML) has shown significant potential in materials science due to its ability to automatically recognize patterns, learn from data, and make informed predictions. In this study, eight ML models were employed to predict the desalination stability of carbon materials. Among these models, the categorical boosting model achieved the highest prediction accuracy. To gain deeper insights, SHapley Additive exPlanations was performed to evaluate the importance of different input features and to identify correlations between these features and desalination stability. Finally, different carbon materials were employed to validate the ML predictions experimentally. The strong agreement between the ML predictions and CDI experimental results underscores the viability of ML in this field. This work pioneers the exploration of ML methods to predict the desalination stability of carbon materials, offering an effective strategy for designing high-stability electrode materials and advancing CDI technology.