Seismic reservoir horizon picking is an active area of research, as it is one of the primary methods of interpretation. A problem with Machine Learning (ML) based methods is that they are prone to being trapped in local minima during optimization, leading to overfitting. This research mainly uses the Support Vector Machines (SVM) method to overcome overfitting and the complex mathematical operations of linear and nonlinear ML methods. We used eleven seismic attributes in our process; there is no need to define a point on the horizon to start hydrocarbon horizon tracking in a local window, which restricts the algorithm to window length and operator choice. Before feeding the input data to SVM, a Principal Component Analysis is performed as a dimensionality-reduction technique to improve efficiency and accuracy. Optimal horizon tracing requires tuning parameters, including regularization control C and the Gaussian kernel width (gamma). In most of the studied cases, the first parameter's effect was negligible for larger gamma values. A mediate-to-coarse Gaussian kernel is more useful for selecting the true horizon locations from multiple. While amplitude information is crucial for feature selection, it is also important to consider complementary information such as phase and frequency-related attributes. Our approach to learning and optimization showcases the robustness and reliability of our workflow for horizon picking. This is evident even in challenging areas with a low signal-to-noise ratio, similar to the outcomes achieved through deep learning in faulted zones, and it allows us to complete tasks in a relatively short amount of time.

Seismic horizon picking, as one of the mainstays of seismic interpretation, has been a topic of discussion in the history of exploration seismology from the age of manual picking methods to the most automated ones. Because of the inherent characteristics of subsurface geologic boundaries and their corresponding sonic features, seismic horizon picking is a multi-dimensional problem. Lateral changes of the corresponding waveform, faults with high throws, unconformities, and signal-to-noise ratio are some of the factors that are responsible for the complexity of the task of horizon picking.The emergence of 3D seismic data acquisition along with complex data processing methods has made seismic data interpretation a time-consuming, labour-intensive and almost manually impossible task to do; therefore, automating procedures was felt necessary by seismic interpreters. Seismic horizon picking was one of the main tasks that have been automated during the last decades using various methods from cross-correlation to the most complicated ones. Machine Learning (ML) based methods have gained popularity for automation, and their ever-increasing development is tempting geophysicists to use them. In our view, ML can be one of the main methods to automate many interpretation procedures because of its inherent non-linearity and high flexibility to any problem that is coincident with the geophysical data.