With the rapid development of technology and algorithms in the last 30 yr, many oil and gas companies now consider machine learning a standard solution in their daily workflow. This paper discusses a case study that uses the random forest algorithm to predict lithofacies of the Balder field from well logs and seismic data. This field is located in the Norwegian sector of the North Sea and has more than 20 yr of production history. The Paleocene reservoirs (Heimdal and Hermod Formations) contain several unconsolidated, remobilized, mound-shaped sandstones. During the sand remobilization, sand injectites were formed within the Lista and Sele Shales, particularly in the Balder Formation, where thick injectites are connected to the remobilized Heimdal and Hermod Sandstones. The three main questions for the research are the following. Which is the best combination of seismic/geometric attributes that can predict the presence of sand bodies? Is it possible to predict the presence of thin sand layers by using this approach? What is the level of confidence in the prediction? The result shows that the random forest models are suitable for predicting thick sand bodies. However, this method struggles to determine the thin sand presence with absolute accuracy compared to the blind-test well data. A new practice is proposed in which cutoff values are applied to improve sand distribution prediction and suggest further research topics to enhance the model results.