According to WHO, approximately 1.71 billion people worldwide have musculoskeletal conditions, which include various issues such as intact muscles, bones, joints, and fractures. Among those, fractures are the most common. Due to the emergency nature of diagnosing fractures, there is a high chance of misdiagnosis for several reasons, such as the unavailability of radiologists, physicians' lack of experience, and other factors. Fracture diagnostic or X-ray interpretation errors can be reduced if radiographs are always read instantly by radiologists or automatically. In our study, we are focusing on automating wrist fracture diagnosis, where we are utilizing the publicly available GRAZPEDWRI-DX dataset, which consists of 20,327 wrist radiographs. We employed the YOLOv9 model for fracture detection, achieving an mAP@50 of 0.677, which surpasses previous benchmarks. For fracture classification, we trained several state-of-the-art deep learning models, including VGG16, VGG19, ResNet50, EfficientNetB7, DenseNet121, MobileNet, and ConvNeXtXLarge. In particular, the YOLOv8-cls model surpassed all others in accuracy (0.93), precision (0.9352), and recall (0.8855), reaching its peak performance at epoch 70. To elucidate the decision-making process of both the detection and classification models, we generated explainable saliency maps through EigenCAM, making the models more explainable and interpretable.