Bone fracture is the most common problem existing in the field of medical emergencies. In Bone fracture detection, conventional approaches typically depend on the expertise of radiologists to interpret radiographic images. However, due to high volume of data and variations among fractures, there is a risk of misinterpretation. Deep learning, particularly convolutional neural networks (CNNs), has transmuted fracture diagnosis through automating this process. In this research, a novel ensemble learning method is introduced, a robust fracture detection model that utilizes an ensemble strategy with added weights to accurately differentiate between normal and fractured bone conditions Through extensive experimentation and validation against few cutting-edge CNN-rooted models adapting transfer learning techniques, the proposed model demonstrates superior performance, exceeding the capabilities of existing methods by a significant margin. A key perspective of this study is the significant impact of base classifier selection on ensemble model performance.