Automated machine learning (AutoML) aims to make high-performance machine learning techniques accessible to a broad audience by automating the design choices in creating machine learning models. This article provides an overview of AutoML's past, present, and future. It introduces the concept of AutoML, defines the problems it addresses, and describes the three key components: search space, search strategy, and performance evaluation. The article discusses hyperparameter optimisation (HPO) techniques, neural architecture search (NAS), and reviews available AutoML systems. It also highlights open challenges and future research directions. AutoML has evolved from early work in algorithm selection and hyperparameter optimisation to more advanced methods like NAS, which automatically designs neural networks. Recent AutoML systems, such as auto-sklearn, TPOT, and AutoKeras, have enabled the creation of high-performance models with minimal user effort. The article discusses the three key components of AutoML: search space (all design choices in a machine learning pipeline), search strategy (methods for finding optimal configurations), and performance evaluation (techniques for assessing model performance). The article also covers benchmarks and meta-learning techniques in AutoML. Benchmarks are essential for evaluating the performance of AutoML systems, while meta-learning helps in warm-starting the search process by recommending initial hyperparameter settings. The article discusses various HPO methods, including random search, grid search, and Bayesian optimisation, and highlights the importance of search space design and transformation. The article reviews the importance of hyperparameters in machine learning models and discusses methods for determining their significance, such as functional ANOVA and ablation analysis. It also explores the transformation of input spaces to improve the efficiency of HPO. The article highlights the challenges in designing search spaces and the importance of considering dependencies between hyperparameters. Overall, the article provides a comprehensive overview of AutoML, covering its key components, techniques, and challenges. It aims to serve as a useful resource for researchers and practitioners in the field of machine learning, offering insights into the current state of AutoML and directions for future research.Automated machine learning (AutoML) aims to make high-performance machine learning techniques accessible to a broad audience by automating the design choices in creating machine learning models. This article provides an overview of AutoML's past, present, and future. It introduces the concept of AutoML, defines the problems it addresses, and describes the three key components: search space, search strategy, and performance evaluation. The article discusses hyperparameter optimisation (HPO) techniques, neural architecture search (NAS), and reviews available AutoML systems. It also highlights open challenges and future research directions. AutoML has evolved from early work in algorithm selection and hyperparameter optimisation to more advanced methods like NAS, which automatically designs neural networks. Recent AutoML systems, such as auto-sklearn, TPOT, and AutoKeras, have enabled the creation of high-performance models with minimal user effort. The article discusses the three key components of AutoML: search space (all design choices in a machine learning pipeline), search strategy (methods for finding optimal configurations), and performance evaluation (techniques for assessing model performance). The article also covers benchmarks and meta-learning techniques in AutoML. Benchmarks are essential for evaluating the performance of AutoML systems, while meta-learning helps in warm-starting the search process by recommending initial hyperparameter settings. The article discusses various HPO methods, including random search, grid search, and Bayesian optimisation, and highlights the importance of search space design and transformation. The article reviews the importance of hyperparameters in machine learning models and discusses methods for determining their significance, such as functional ANOVA and ablation analysis. It also explores the transformation of input spaces to improve the efficiency of HPO. The article highlights the challenges in designing search spaces and the importance of considering dependencies between hyperparameters. Overall, the article provides a comprehensive overview of AutoML, covering its key components, techniques, and challenges. It aims to serve as a useful resource for researchers and practitioners in the field of machine learning, offering insights into the current state of AutoML and directions for future research.