Active Learning (AL) is a prominent research area in machine learning that focuses on reducing labeling costs by strategically selecting the most informative or uncertain data points for annotation, thereby improving model performance with fewer labeled examples. Foundational approaches include uncertainty sampling, query-by-committee, and expected model change, which guide the selection of instances that maximize learning efficiency. Recent research extends these methods to deep learning, leveraging techniques such as Bayesian neural networks, reinforcement learning-based query strategies, core-set selection, and adversarial active learning to handle high-dimensional and complex data. Applications of active learning span natural language processing, computer vision, speech recognition, medical imaging, and autonomous systems, where acquiring labeled data is expensive or time-consuming. Current studies also explore batch-mode selection, multi-label active learning, and integration with semi-supervised and federated learning, establishing AL as a key approach for building data-efficient, robust, and cost-effective AI systems.