The Synthetic Minority Oversampling Technique (SMOTE) is a widely used preprocessing method for handling imbalanced data. Since its introduction in 2002, SMOTE has become a standard in the field, known for its simplicity and effectiveness in generating synthetic minority class examples. It works by interpolating between neighboring minority class instances to create new examples, thereby improving the classifier's ability to generalize. SMOTE has inspired numerous approaches to address class imbalance and has contributed to various supervised learning paradigms, including multilabel classification, incremental learning, and semi-supervised learning. Over the past 15 years, SMOTE has been extended and adapted to various learning scenarios, including streaming data, semi-supervised learning, multi-instance learning, and regression. These extensions aim to enhance SMOTE's performance in different contexts, such as handling high-dimensional data, dealing with overlapping classes, and addressing the challenges of big data. However, SMOTE still faces several challenges, including the need to handle small disjuncts, noisy data, and data shifts between training and test sets. Additionally, SMOTE's effectiveness in real-time data streams and large-scale datasets remains a topic of ongoing research. This paper reviews the development and impact of SMOTE over the past 15 years, highlights its contributions to machine learning and data mining, and identifies future challenges in extending SMOTE for big data problems. It also discusses various SMOTE-based extensions and their applications in different learning paradigms, emphasizing the importance of addressing the unique challenges posed by imbalanced data. The paper concludes with a summary of the key findings and future directions for research in the field of imbalanced learning.The Synthetic Minority Oversampling Technique (SMOTE) is a widely used preprocessing method for handling imbalanced data. Since its introduction in 2002, SMOTE has become a standard in the field, known for its simplicity and effectiveness in generating synthetic minority class examples. It works by interpolating between neighboring minority class instances to create new examples, thereby improving the classifier's ability to generalize. SMOTE has inspired numerous approaches to address class imbalance and has contributed to various supervised learning paradigms, including multilabel classification, incremental learning, and semi-supervised learning. Over the past 15 years, SMOTE has been extended and adapted to various learning scenarios, including streaming data, semi-supervised learning, multi-instance learning, and regression. These extensions aim to enhance SMOTE's performance in different contexts, such as handling high-dimensional data, dealing with overlapping classes, and addressing the challenges of big data. However, SMOTE still faces several challenges, including the need to handle small disjuncts, noisy data, and data shifts between training and test sets. Additionally, SMOTE's effectiveness in real-time data streams and large-scale datasets remains a topic of ongoing research. This paper reviews the development and impact of SMOTE over the past 15 years, highlights its contributions to machine learning and data mining, and identifies future challenges in extending SMOTE for big data problems. It also discusses various SMOTE-based extensions and their applications in different learning paradigms, emphasizing the importance of addressing the unique challenges posed by imbalanced data. The paper concludes with a summary of the key findings and future directions for research in the field of imbalanced learning.