This study introduces a methodology for the improvement of the visibility of archaeological features using an open-source probabilistic machine learning framework applied to UAV LiDAR data from the Torre Castiglione site in Apulia, Italy. By leveraging a Random Forest classification algorithm embedded in an open-source software, the approach processes dense LiDAR point clouds to segment out vegetation from the ground and the structures. Key steps include training the classifier, generating digital terrain models, digital feature models, and digital surface models, and enhancing the visibility of archaeological features. This method has proven effective in improving the interpretation of archaeological sites, revealing previously hidden or difficult-to-access microtopographic and structural details, such as the defensive structures, terraces, and ancient paths of the Torre Castiglione site. The results underline this methodology’s ease of use in uncovering archaeological landscapes under a dense canopy. Moreover, the study emphasises the benefits of using open-source tools to enhance the documentation and analysis of remote or difficult archaeological sites.