A Structure-Property Screening Framework for Polymer Shell Encapsulation of Phase-Change Materials: Random Forest and...

Confining a phase-change material (PCM) within a polymer shell yields leak-proof, mechanically robust latent-heat storage media, but selecting a shell chemistry and polymerization route requires balancing competing targets: latent-heat storage density (ΔH, the melting enthalpy per unit capsule mass), core loading content (LC), capsule diameter (d), and a melting temperature (T_m) matched to the application. Because the literature characterizes each method-shell-core combination in isolation, these structure-property relationships cannot be compared quantitatively across studies. We present a proof-of-concept, data-driven framework linking shell and process descriptors to encapsulation performance. From a curated dataset of 90 micro- and nano-encapsulated PCM records (53 with measured ΔH) spanning 11 encapsulation routes and eight shell material families, Random Forest (RF) and Gaussian Process (GP) surrogates predict ΔH, and a non-dominated sorting genetic algorithm (NSGA-II) optimizes ΔH, LC, and d over the continuous (T_m, LC) space for every method-shell-core trio with at least three records (n = 11). Benchmarked against mean, linear-LC, and physics-informed baselines under repeated cross-validation, the surrogates match but do not exceed the elementary baselines (median R² ≈ 0.33), a result we report honestly given the modest sample size. The Matérn GP provides borderline-calibrated uncertainty, supporting a robust, extrapolation-penalizing NSGA-II. Hypervolume rankings place emulsion polymerization, sol-gel silica, and in situ polymerization as the top-performing methods under both nominal and robust criteria. Presented as a methodology demonstration rather than a definitive ranking, the framework, with full code and data, is a reusable approach for structure-property quantification of polymer-encapsulated PCMs as experimental data accumulate.