The use of wireline facies associations can alleviate core data shortage during facies prediction by providing a more extensive input dataset. Wintershall has assigned wireline facies associations directly on cored and un-cored wells in the Carboniferous of the Sothern North Sea. Conducting facies prediction using these wireline facies associations as an input can help with tapping into the remaining exploration and development potential of the area. However, the accuracy of this input must be evaluated using core data before machine learning algorithms are applied. This was quantified as 71% for 9 cored wells, where the background floodplain and braided channel facies had the highest accuracies of 88% and 81% respectively, and the mouth bars and marine shales facies could not be adequately validated due to their insufficient core sampling. Consequently, when using wireline facies associations for training facies prediction algorithms, this input’s intrinsic uncertainty should be accounted for while examining the outputs, especially for facies that are not sufficiently validated by cores. Applying facies prediction with Support Vector Machine (SVM), Multilayer Perceptron Neural Network (MLP) and Recurrent Neural Network (RNN), showed that RNN can achieve the highest overall accuracy of 80.9%, due to the highest F1 scores for braided channel (0.88), point bars (0.60) and coal (0.53). The class imbalance problem is apparent for this dataset where the majority classes of background floodplain, braided channel, point bar and coal, are more predicted than the minority classes of crevasse splay sands, mouth bars, and marine shale. Applying RNN on the Westphalian A, B and C separately served as a form of imbalance correcting technique that increased the F1 scores of underrepresented facies. Future work can further refine the results by exploring imbalance correcting techniques through under-sampling the background floodplain and over-sampling the crevasse splay, mouth bar and marine shale facies.