<p>Lithofacies are critical to understanding reservoir characteristics, and their accurate identification is essential for optimizing strategies for development in shale reservoirs. However, predicting the vertical and lateral continuity of lithofacies using conventional logging data is often limited by formation heterogeneity and the complexity in lithofacies classification. In this study, we focus on the W Formation in the western margin of the Ordos Basin, where lithofacies were classified based on lithology, sedimentary structures, and brittleness. This was achieved by employing an integrated approach involving core observations, low-pressure CO₂ and N₂ adsorption tests, high-pressure mercury injection, borehole electrical image logging, and Litho Scanner logging data. To predict lithofacies, we employed multi-resolution graph-based clustering (MRGC), hierarchical ascendant clustering (HAC), and dynamic clustering (DC) algorithms using information from conventional logging data. Our findings reveal that layered limestones and tuffs exhibit favorable reservoir properties, leading us to classify lithofacies into three distinct types: shale facies, layered shale facies, and limestone facies. Notably, the layered shale facies demonstrated superior porosity, total organic carbon content, and brittleness, rendering them “sweet spots” for reservoir potential. All three clustering algorithms yielded promising results, with prediction accuracies exceeding 75% for each lithofacies. The DC algorithm performed particularly well, achieving over 80% accuracy across all lithofacies types, and a remarkable 91.99% accuracy for layered shale facies. These predicted facies closely correspond with the highly productive layers identified through fracturing tests. The lithofacies classification and predictive model developed in this study provide a valuable framework for identifying favorable lithofacies distributions and optimizing shale reservoir development strategies.</p>