Unsupervised Electrofacies Classification and Porosity Characterization in the Offshore Keta Basin Using Wireline Logs

A team of researchers from Ghana and South Africa, including Hamdiya Adams and Theophilus Ansah-Narh, developed an unsupervised machine learning workflow for electrofacies classification in the offshore Keta Basin, where conventional core data is limited. The study, accepted at ICECET 2026, analyzed six standard wireline logs from Well C over a depth interval covering approximately 11,195 samples. Using K-means clustering in multivariate log space, the team evaluated clustering quality with inertia and silhouette diagnostics, achieving an average silhouette coefficient of about 0.50—indicating moderate but meaningful separation. Four distinct clusters emerged, corresponding to a geological continuum from clay-rich shales to cleaner sandstone-dominated units. The electrofacies exhibit systematic, depth-continuous patterns tied to variations in clay content, porosity, and rock framework properties.

This log-only clustering framework, validated by quantitative metrics, offers a robust and reproducible method for subsurface characterization in frontier offshore basins. By eliminating the need for extensive core sampling, it significantly reduces exploration costs and timelines. The proposed workflow is positioned as a practical tool for early-stage formation evaluation, enabling rapid identification of reservoir-quality intervals. The authors note that the approach can serve as a foundation for future integrated studies, potentially incorporating additional logs or more advanced clustering algorithms. This work highlights the growing role of AI in geoscience, particularly for resource-constrained regions like the West African margin.