Integrated Supervised and Unsupervised Machine Learning Approach to Map the Electrochemical Windows Over 4500 Solvents for Battery Applications.

The compatibility between solvent electrolytes and high-voltage electrode materials represents a significant impediment to the progress of rechargeable metal-ion batteries. Rapidly identifying suitable solvent electrolytes with optimized electrochemical windows (ECWs) within an extensive search space poses a formidable challenge. In this study, we introduce a combined supervised and unsupervised (clustering) machine learning (ML) approach to discern distinct clusters of solvent electrolytes exhibiting varying ECW ranges. Through supervised machine learning, we have accurately predicted optimal solvent electrolytes with desired ECWs from a vast pool of 4882 solvents. Our ML model boasts superior accuracy compared to previously reported data from density functional theory (DFT). Besides, the exploration of the vast solvent space through K-means clustering (unsupervised approach) yields 11 optimal clusters, each encompassing different solvents characterized by diverse ECW ranges and frequencies. The expedited reduction of solvent space achieved through clustering occurs within a very short time frame and with minimal resource expenditure. Consequently, this method is highly capable of streamlining the subsequent experimental investigations for battery applications, avoiding the need for a trial-and-error approach.