Capturing small-scale features in complex subsurface geology, such as Carbonate, through seismic imaging poses challenges due to the influence of heterogeneous properties of objects in the subsurface on propagated waves. The first step in machine learning (ML) involves supplying a sufficient amount of data to ensure the learning algorithm is updated and matured. In the absence of multiple shapes of diffraction data, the accuracy of your prediction for machine learning (ML) may be compromised. ML may also fail to detect the pattern of diffraction in the data. Following the learning process, our machine focuses on the critical task of target detection. This involves comparing the target with the given data and searching for a specific signature. In this study, data is provided to the system in the form of images and features. The learning algorithm can process the input to make predictions about the target. The concept behind machine learning is to minimize the discrepancy between your prediction and the target value as much as feasible. The preservation of diffraction amplitude in laterally varying velocity conditions is influenced by certain factors. The technique of ML destruction is employed to separate diffraction data, as traditional filtering methods tend to blend diffraction amplitudes in cases where there are one or multiple diffractions present. Copyright © 2024, International Petroleum Technology Conference.