Lithium-metal batteries (LMBs) have much higher energy densities than the current state-of-the-art lithium-ion batteries that drive the green energy landscape. However, their wide-scale commercialization is hampered by the lack of desirable electrolytes that suffer from stability and cyclability issues. The discovery of electrolytes for LMBs has been mainly guided by chemical intuition and trial-and-error experiments. While the emerging machine learning (ML)-based initiatives for electrolyte discovery are promising, most studies have been carried out on limited datasets of ionic conductivity, they remain experimentally unvalidated, and lack in design principles governing performance of efficient electrolytes. The present work addresses these problems by curating exhaustive datasets of the most critical parameters affecting the LMB electrolytes – ionic conductive, oxidative stability, and Coulombic efficiency and by developing highly accurate ML and deep learning (DL) models. The models were consistent with the known chemical principles of ionic conductivity and pinpointed few non-trivial trends. After undergoing stringent checks on out-of-distribution (OOD) in-house data sets for the three target properties, the ML models were finally deployed on large unlabeled datasets to identify new and promising electrolytes, and a score was devised to rank the molecules in terms of the predicted target properties. The subsequent experimental investigations led to an entirely new class of LMB electrolytes, establishing the efficacy of our heuristic approach. The proposed “multi-pronged” strategy for LMB electrolyte discovery will also enable researchers to find extraordinary electrolytes for other battery chemistries.