Integrating multiple seismic attributes for fault detection using a new hybrid machine learning

This paper proposes a novel hybrid approach for fault detection in seismic data by integrating multiple seismic features using a novel hybrid machine. This hybrid method introduces the integration of Multilayer Perceptron (MLP) neural networks with Support Vector Machines (SVM). The main objective of this study is to enhance the analysis and recognition of fault patterns in seismic data while reducing detection errors. The dataset consists of two-dimensional synthetic and real seismic data and their corresponding labels. Various seismic features, such as the gray-level co-occurrence matrix (GLCM), and features derived from ant tracking, chaos, variance, sweetness, correlation, slope direction, and energy, are extracted and normalized. These features are subsequently used to train MLP and SVM models for fault detection. SVM, a supervised learning method, operates by determining a hyperplane that maximizes the separation between data classes. In contrast, MLP, an artificial neural network, uses multiple layers to optimize weights and capture complex data relationships. The performance of each model is evaluated using experimental data, and its corresponding accuracies are calculated. Finally, the predictions from both models are combined to improve the overall fault detection accuracy. The results show that this combined approach significantly increases the fault detection accuracy compared to the independent application of each model.