Integrating phase change materials (PCMs) in buildings cannot only enhance the energy performance, but also improve the renewable utilization efficiency through considerable latent heat during charging/discharging cycles. However, system performances are dependent on PCMs’ integrated forms, heat transfer enhancement solutions, system operating modes, together with optimal geometrical and operating parameters. In this study, passive, active, and combined passive/active solutions in PCMs systems have been comprehensively reviewed, when being applied in heating, cooling and electrical systems, together with a dialectical analysis on advantages and disadvantages. In addition to novel system designs, interdisciplinary applications of machine learning have been reviewed and formulated, from perspectives of reliable structures, smart operational controls, and stochastic uncertainty-based performance prediction. Furthermore, a generic methodology with a systematic and hierarchical procedure has been proposed, with the implementation of machine-learning based technique for optimisations during both design and operation periods. The mechanisms of machine learning techniques were characterised as the simplifications of modelling and optimization processes, through the errors-driven update, the support vector regression and the backpropagation neural network. Several technical challenges were identified, such as the heat transfer enhancement, the novel structural configurations and the flexible switch on operating modes. Finally, identified challenges on machine learning include the development of advanced learning algorithms for efficient performance predictions, optimal structural configurations on neural networks, the trade-off between computational complexity and reliable optimal solutions, and so on. The formulated climate-adaptive designs, intelligent operations, uncertainty-based analysis and optimisations with interdisciplinary machine learning techniques can promote PCMs applications in sustainable buildings.@en