K-means clustering using principal component analysis to automate label organization in multi-attribute seismic facies analysis

The use of unsupervised machine learning methods such as K-means, Hierarchical Agglomerative Clustering, and Self-organizing maps is constantly increasing in seismic interpretation. Regarding unsupervised methods, the K-means technique is one of the simplest ways to cluster seismic facies, although it presents neither a structure between the generated labels nor a measure of similarity when considering their order. To solve this drawback, we propose two automated label organization techniques that use principal component analysis (PCA) to organize those obtained from the algorithm, preserving some degree of similarity. To demonstrate the effectiveness of these methods, we interpreted two stratigraphic surfaces known as Maximum Transgressive Surface (MTS) and Maximum Regressive Surface (MRS), then extracted some attributes to run clustering experiments. Furthermore, we performed the principal component analysis and selected the first three components to be clustered. Subsequently, these components were used to organize the labels obtained with K-means through the two proposed techniques. Finally, we interpreted the outstanding results obtained from the methodologies proposed, allowing us to understand better seismic facies and the depositional environments over stratigraphic surfaces.