Application of computational intelligence tools for the analysis of marine geotechnical properties in the head of Zakynthos canyon, Greece

This paper uses a computational approach to provide insight into the relationships among marine geotechnical properties that characterize the recent sedimentary cover at the head of Zakynthos Canyon in western Greece. Self-organizing maps (SOM) and generic interaction matrix (GIM) theory were used to investigate the tendency of the data to cluster and to examine the sediment property relationships. This analysis has also focused on the assessment of the dominance and interaction intensity between the related parameters following GIM theory definition. The principal results refer to the identification of clusters in the original multivariate data set. SOM-based analysis distinguished five clusters, with similar geotechnical characteristics, which led to the separation of the surficial (∼80 cm) unconsolidated sediments from the deeper normally consolidated sediments and depicted better relations between the geotechnical properties within each cluster. The combination of SOM with GIM theory also demonstrates the dominance of fine-grained sediments (especially silts) and their associated Atterberg limits. The strongest interaction intensity is observed between silt and water content, whereas the undrained shear strength of the surficial deposits appears to be least interactive. The application of computational intelligence methods in the study of marine geotechnical properties allows insight into the relationships between the various geotechnical parameters and provides a promising tool for knowledge extraction in marine geo-environments.

► We examine the relationships amongst marine geotechnical properties.
► We apply computational intelligence tools to investigate data tendency to cluster.
► We analyze the dominance and interaction intensity between the related parameters.
► We conclude that clustering is generated according to certain burial depth interval.
► Fine-grained sediments seem to play an important role dominating the data set.