Ordos Basin's Chang 7 Member of Yanchang Formation, as a typical continental shale oil and gas reservoir, exhibits widespread development of mud shale characterized by low calcium mineral content and high content of siliceous and clay minerals. In this study, lithofacies types were classified based on conventional well log curves, whole rock mineral analysis, and organic matter content. A total of 4 one-level lithofacies and 16 two-level lithofacies were identified. Specifically, understanding of the mud shale lithofacies in the Chang 7 Member was enhanced, revealing three main lithofacies: calcium clay siliceous shale, calcium-containing siliceous/clay mixed shale, and calcium-containing siliceous clay shale. Due to the high cost and low popularity of experiments such as whole rock mineral determination, this study employed a multilayer perceptron neural network to establish a connection between conventional well log curves and whole rock mineral content to predict the whole rock mineral content in the study area. Meanwhile, cluster analysis and discriminant analysis were used to autonomously classify lithofacies types in the study area. The results indicate that the combination model of multilayer perceptron neural network and Bayesian discriminant analysis has a high predictive accuracy for lithofacies, reaching 89.9%, making it a major method for lithofacies prediction.