Machine Learning for Ionic Conductivity Prediction in Polymer Electrolytes: A Benchmark on Lithium Systems and a Feasibility Study on the Data-Scarce Multivalent (Mg, Zn, Na) Frontier
Year: 2026
Type: preprint
Abstract: Solid and gel polymer electrolytes are central to safer, more flexible batteries, and machine learning has been used to predict their ionic conductivity directly from composition and processing variables. Almost all of this work targets lithium-ion systems, for which large, curated datasets exist. Polymer electrolytes conducting multivalent or non-lithium cations (magnesium, zinc, sodium) are promising lower-cost alternatives but remain comparably underexplored, in part because no public, ML-ready dataset of their measured ionic conductivities exists. Here we validate a standard composition-to-conductivity pipeline (Random Forest and XGBoost regression) on OBELiX, a published dataset of 599 lithium solid-state electrolytes, reaching R2 = 0.79 against a mean-predicting baseline near zero. We then construct, to our knowledge, the first ML-ready dataset of Mg-, Zn-, and Na-conducting polymer electrolytes, systematically mined from the primary and review literature: 135 data points across 105 distinct host-polymer/salt systems. Applying the same pipeline to this new dataset, a tuned Random Forest reaches R2 = 0.47 under leave-one-out cross-validation, or 0.36 under a more conservative 5-fold estimate, well below the lithium benchmark. Through chemistry-informed feature engineering and a nested cross-validation check, we show this gap is driven primarily by data scarcity and literature measurement heterogeneity, not by model or feature choice. We argue the scarcity is itself the central finding: it puts a number, for the first time, on how far behind the multivalent and non-lithium polymer electrolyte literature sits relative to the lithium-ion mainstream. We release the curated dataset and pipeline to support future community data collection.
Source: ChemRxiv
Author: Meghanadh Singh Yadav Dokku
Language: English
Subfield: Materials Chemistry
Field: Materials Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7223137897