Machine learning-based prediction of microstructural evolution during heat treatment of high-entropy alloys: A comprehensive review

High-entropy alloys (HEAs) are multi-principal-element material systems in which five or more elements are combined in approximately equimolar proportions. They have attracted sustained scientific attention because they can unite mechanical, thermal, and chemical properties that surpass those of conventional single-principal-element alloys. Heat treatment is a principal lever for tuning HEA microstructures: annealing, homogenisation, ageing, and solution treatment each alter phase fractions, precipitate morphologies, grain size distributions, and dislocation substructures in ways that ultimately determine suitability for demanding aerospace, energy-generation, or biomedical service conditions. The experimental exploration of this enormous compositional and processing space is, however, prohibitively expensive and time-consuming when pursued through conventional trial-and-error metallurgy alone. The emerging closed-loop paradigm, in which ML models predict microstructural outcomes, active learning selects the next experiment, and automated characterisation closes the feedback loop, represents the critical frontier of HEA heat-treatment science, enabling systematic navigation of composition-processing-microstructure space at throughputs unachievable by purely experimental campaigns. Machine learning (ML) methods including random forests, gradient-boosted trees, artificial neural networks, support vector machines, Gaussian process regressors, graph neural networks, convolutional neural networks, long short-term memory networks, and physics-informed deep learning architectures, offer a fundamentally different paradigm: they extract quantitative structure-property-processing relationships directly from accumulated datasets, then generalise those relationships to compositions and heat treatment schedules that have never been physically tested. This review synthesises the state of the art in ML-assisted prediction of microstructural evolution during the heat treatment of HEAs. We trace the foundational metallurgical concepts governing HEA phase selection and microstructural development, survey the construction and curation of datasets on which ML models are trained, evaluate principal algorithms and their relative merits and shortcomings, and discuss the integration of ML with CALPHAD thermodynamic frameworks and physics-informed constraints. We then identify critical open challenges such as data scarcity, transferability, interpretability, and the treatment of kinetic pathways and propose research directions expected to define the next decade of work in this field. The scope of this review is deliberately bounded: it covers ML methods applied to heat-treatment-induced microstructural evolution, including phase fraction, precipitate morphology, grain size, and sigma-phase formation in HEAs and directly related MPEA systems; coverage of room-temperature mechanical property prediction, corrosion, and coating deposition is included only where it directly informs microstructural evolution modelling. Although Section 10 gives particular attention to surface engineering and high-entropy coating applications, the methods reviewed are equally applicable to bulk structural alloys for aerospace and nuclear service, biomedical implant alloys, hydrogen-storage hydrides, thermoelectric devices, and energy-storage electrodes; this breadth is addressed in Section 10.4. The review encompasses 106 verifiable primary references distributed across 12 sections and 15 original figures.