X-ray images are widely used to identify issues in bones and various human organs. However, in certain situations—particularly when a radiologist's experience is limited or during emergency medical services where immediate consultation with a radiologist is unavailable—patients may receive incorrect treatments due to the potential for inaccurate diagnosis of suspected fractures. Therefore, the initial phase of treatment, which involves detecting fractures or fissures, is critically important. This research aims to facilitate accurate and prompt fracture diagnosis for clinicians, especially in forensic scenarios or medical settings with limited access to experienced radiologists or orthopedists, thereby improving clinician performance and patient care. We evaluated the efficiency of the EfficientNet architecture, a convolutional neural network model, for classifying bone fractures. We created a deep learning application aimed at detecting fractures in different bones, such as the elbow, finger, forearm, hand, humerus, shoulder, and wrist. This application was developed using the MURA dataset, with the goal of being implemented in clinical settings. The performance of the proposed approach was assessed against the MURA dataset and original data obtained from a hospital. The system demonstrated significantly better performance on the hospital dataset compared to MURA. The average recognition rates of the proposed approach were 80.3% for the finger, 87.9% for the forearm, and 96.2% for the hospital dataset.