Potholes are a widespread and bothersome road fault that provide a considerable barrier to infrastructure maintenance and driving safety globally. In addition to putting vehicles in danger, these road surface depressions cause significant expenses for the government and private citizens in the form of car damage, accidents, and repair bills. In order to detect, this study compares transfer learning and machine learning (ML) models. The dataset was created by gathering photos from various sources, including automobiles, maps, and Android smartphones. To obtain comparable features, a non-overlapping two-second moving window was used during the preprocessing stage. According to the paper, the transfer learning model was created with Tensorflow, CNN, and MobileNet. The algorithm outperforms the other approaches with an accuracy of 95.3%, precision of 97.1%, f1-score of 93.4%, and recall of 93.2% over the training dataset.