Unsupervised boundary analysis of potential field data: A machine learning method

We developed a method for interpretation of potential field data through unsupervised machine learning techniques, called Unsupervised Boundary Analysis (UBA). Through the use of Self-Organizing Maps and k-means it is possible to obtain a clustering in the space domain allowing the identification of the edges of the sources, with no prior information and no need of expert interpreters. To understand the behavior of our procedure we performed a test on gravimetric synthetic data of two vertical faults. Then, we tested UBA on the real case of Torre Galli (Calabria, Italy), highlighting several buried archaeological bodies. The results were also compared with two boundary analysis other methods.