Inversion of magnetic data to characterisegeological structures, such as dikes, is a fundamental challenge in engineering geophysics due to its highly non-linear and ill-posed nature, necessitating robust optimization methods. This study introduces and evaluates for the first time, the Mountain Gazelle Optimizer (MGO) for the first time, examining its efficiency and potential as an effective solution to this problem. The MGOalgorithm is designed to find the global optimum by intelligently balancing exploration and exploitation within the parameter space. The performance of the MGO was assessed by comparing it with two distinct approaches: a powerful machine learning algorithm called Random Forest (RF), and a classic processing-estimation method based on Reduction to the Pole (RTP). Evaluations were conducted on synthetic data (with noise levels ranging from 0% to 20%) as well as on real field data from the Gansu iron deposit in China. The results clearly demonstrated the superiority of MGO in all scenarios. Not only did the algorithm exhibit greater stability against noise than RF, it also, achieved a Root Mean Square Error (RMSE) of 0.48 in the real data case study,, which was significantly lower than the error achieved by the classic method (0.88). Furthermore, the parameters estimated by MGO showed better alignment with the geological information from existing drilling data in the area. This study suggests that MGO's superiority obtained from its direct and global inversion approach. Ultimately, MGO is presented as an accurate and reliable tool for exploration and engineering applications.
