This work presents three automated pre-trained models to predict the difficulty of extracting the mandibular third molar using a dataset of 2414 panoramic radiography images based on pre-processed (shifted and rotated) from left and right mandibular third molar instances. In this research, we employed the four distinct architectural models, namely VGG-16, VGG-19, MobileNetV2, and ResNet50 to identify the difficulty of removing a mandibular third molar. We categorized the dataset into four categories of complexity to help in categorization (Normal, Easy, Medium, and difficult). As a result, VGG-16, VGG-19, MobileNetV2 and ResNet50 had prediction accuracies of 81%, 82%, 79% and 44%, respectively. Findings affirmed that the proposed deep learning model using VGG-19 could be a good tool to predict the difficulty of extracting a mandibular third molar using a panoramic radiographic image.