Many topics make large-scale labeled dataset acquisition impossible due to expensive annotation costs, privacy concerns, or data scarcity. Active Learning (AL) frameworks offer an effective alternative by proactively picking the most informative examples for labeling, resulting in maximum model performance with minimal supervision. This study investigates the implementation of intelligent sampling algorithms in small-scale data settings. To maximize sample selection, we propose a hybrid framework that integrates criteria for uncertainty, diversity, and representativeness. Experiments on benchmark small datasets show that our strategy greatly decreases labeling labor while maintaining competitive accuracy compared to existing sampling methods. Our findings emphasize the ability of intelligent AL systems to democratize machine learning in resource-constrained environments