Inversion of Magnetic Anomaly using Machine Learning Regression Techniques along with PSO

Summary In this study, we have compared the Particle Swarm Optimization (PSO) algorithm, Artificial Neural Network, K- nearest Neighbors, Random forest regressor for the inversion of magnetic anomaly dataset. We have tried to find the unknown parameters of the causative source i.e., the horizontal position of the source, depth, amplitude coefficient, angle of effective magnetization, and the shape factor. Machine learning algorithms are implemented by constraining shape factors obtained from PSO. The applicability of these algorithms is tested both for noise-free and 10% gaussian noisy synthetic magnetic anomaly data generated for simple geometrical bodies like a sphere, horizontal cylinder and thin dyke. In addition, the validity of algorithms is also demonstrated on two field examples namely Bankura Anomaly, India, and Pima copper deposit, Arizona USA. Inversion parameters obtained from these PSO and ML algorithms are well corroborated with the results from previous studies.