Recent advances and applications of machine learning in solid-state materials science Machine learning has become a powerful tool in materials science, significantly accelerating both fundamental and applied research. This review provides a comprehensive overview of recent developments and applications of machine learning in solid-state materials science. It introduces machine learning principles, algorithms, descriptors, and databases in materials science. The review discusses various machine learning approaches for discovering stable materials and predicting their crystal structures, as well as research on quantitative structure-property relationships and replacing first-principle methods with machine learning. It also explores how active learning and surrogate-based optimization can improve the rational design process and provides examples of applications. The review addresses the interpretability of machine learning models and their physical understanding, and proposes solutions and future research paths for challenges in computational materials science. Machine learning algorithms have revolutionized other fields, such as image recognition. However, the development of machine learning from the first perceptron to modern deep convolutional neural networks was a long process. In materials science, machine learning is often used for supervised learning, which requires labeled data. The success of machine learning in materials science depends on the amount and quality of data available, which is a major challenge. Databases such as the Materials Project, the Inorganic Crystal Structure Database, and others are essential for materials informatics. Machine learning in materials science involves the representation of data in a suitable form, with features capturing all relevant information. Features can be simple, such as atomic numbers, or complex, such as expansions of radial distribution functions. The process of feature extraction or engineering is crucial for the success of machine learning models. The choice of features depends on the target quantity and the variety of the space of occurrences. Various descriptors, algorithms, and databases are used in materials informatics. Databases such as the Materials Project, the Inorganic Crystal Structure Database, and others are essential for materials informatics. A FAIR treatment of data is required for these databases to thrive. Negative results are often discarded, but they are as important for machine learning algorithms as positive results. In some disciplines, such as chemistry, databases already exist for negative results. Features in materials science must be able to capture all relevant information, necessary to distinguish between different atomic or crystal environments. The process of feature extraction or engineering can be as simple as determining atomic numbers or involve complex transformations. The best choice for the representation depends on the target quantity and the variety of the space of occurrences. Various descriptors, such as Coulomb matrices, Weyl matrices, Z-matrices, and others, are used to represent the local structural environment. These descriptors are crucial for the success of machine learning models in materials science. The SOAP kernel is one of the most successful atomic environment features, capable of characterizing the entire atomic environment at once. In addition, various algorithms, such as ridge regression, kernel ridge regression, and decision tree-based methods like random forests and extremely randomized trees, are used in materials science. These algorithms are essential for theRecent advances and applications of machine learning in solid-state materials science Machine learning has become a powerful tool in materials science, significantly accelerating both fundamental and applied research. This review provides a comprehensive overview of recent developments and applications of machine learning in solid-state materials science. It introduces machine learning principles, algorithms, descriptors, and databases in materials science. The review discusses various machine learning approaches for discovering stable materials and predicting their crystal structures, as well as research on quantitative structure-property relationships and replacing first-principle methods with machine learning. It also explores how active learning and surrogate-based optimization can improve the rational design process and provides examples of applications. The review addresses the interpretability of machine learning models and their physical understanding, and proposes solutions and future research paths for challenges in computational materials science. Machine learning algorithms have revolutionized other fields, such as image recognition. However, the development of machine learning from the first perceptron to modern deep convolutional neural networks was a long process. In materials science, machine learning is often used for supervised learning, which requires labeled data. The success of machine learning in materials science depends on the amount and quality of data available, which is a major challenge. Databases such as the Materials Project, the Inorganic Crystal Structure Database, and others are essential for materials informatics. Machine learning in materials science involves the representation of data in a suitable form, with features capturing all relevant information. Features can be simple, such as atomic numbers, or complex, such as expansions of radial distribution functions. The process of feature extraction or engineering is crucial for the success of machine learning models. The choice of features depends on the target quantity and the variety of the space of occurrences. Various descriptors, algorithms, and databases are used in materials informatics. Databases such as the Materials Project, the Inorganic Crystal Structure Database, and others are essential for materials informatics. A FAIR treatment of data is required for these databases to thrive. Negative results are often discarded, but they are as important for machine learning algorithms as positive results. In some disciplines, such as chemistry, databases already exist for negative results. Features in materials science must be able to capture all relevant information, necessary to distinguish between different atomic or crystal environments. The process of feature extraction or engineering can be as simple as determining atomic numbers or involve complex transformations. The best choice for the representation depends on the target quantity and the variety of the space of occurrences. Various descriptors, such as Coulomb matrices, Weyl matrices, Z-matrices, and others, are used to represent the local structural environment. These descriptors are crucial for the success of machine learning models in materials science. The SOAP kernel is one of the most successful atomic environment features, capable of characterizing the entire atomic environment at once. In addition, various algorithms, such as ridge regression, kernel ridge regression, and decision tree-based methods like random forests and extremely randomized trees, are used in materials science. These algorithms are essential for the