A Review on Use of Machine Learning Techniques in Predictive Selection of Phase Change Material

Abstract

Phase Change Materials (PCMs) are central to thermal energy storage systems, but selecting the most suitable material remains difficult because thermal performance depends on many interacting properties. Machine learning (ML) offers a practical route for predictive selection by learning relationships among composition, structure, thermophysical properties, and performance indicators. This review summarizes current research on ML-based prediction and optimization for PCM selection, with emphasis on melting temperature, latent heat, thermal conductivity, density, stability, and cycling reliability. Recent studies show that regression, ensemble learning, artificial neural networks, and hybrid AI workflows can reduce experimental effort while improving screening speed and model accuracy. The literature also indicates that data quality, feature engineering, cross-study comparability of reported accuracy metrics, and interpretability remain major challenges for adoption in materials design; this review cross-checks representative quantitative claims against their original sources and reports the resulting corrections explicitly. Based on the reviewed evidence, ML is not replacing experiments; rather, it is complementing them by narrowing the candidate pool and guiding targeted validation. The paper concludes that future PCM discovery will likely depend on integrated datasets, explainable models, and domain-aware optimization frameworks that link material chemistry with storage performance.