Hydrothermal liquefaction (HTL) has potential for converting the world’s massive quantities of wet organic wastes into renewable fuels. Wet wastes fed to HTL undergo a complex series of reactions, resulting in formation of an energy-dense biocrude product. The biocrude yield is one of the most important parameters determining economic viability, making its prediction for a given feedstock an important capability for allocating finite resources for investment in HTL commercialization. The complex reaction network, consisting of hundreds of reactants and thousands of reactions, is a challenge for traditional, physics-based predictions of reaction outcomes. Data-driven techniques constitute a newly emerging alternative to physics-based prediction methods. Ensuring the accuracy of data-driven models requires careful data curation, model selection, and method development. In this work, a data set was assembled consisting of 570 data points that had previously been published in the open literature After dividing the data set into training and testing sub-sets, eight different algorithms were tested for accuracy, using seven different compositional properties as the independent variables. Among the tested algorithms, Random Forest and eXtreme Gradient Boosting (XGBoost) made the most accurate predictions of biocrude yield compared with the test set, with root mean square errors (RMSE) of 8.34 and 8.57, respectively. Further refinement of the Random Forest model reduced its RMSE to 8.07. Predictions made by a series of literature models for the same test set data resulted in RMSE ranging from 9.2 in the most accurate case to 27.6. Finally, eight feeds that have never been studied previously using HTL were selected from the literature and biocrude yields predictions were made for them. The biocrude yields and corresponding uncertainties were fed into a probabilistic economic model to project the minimum fuel selling price for a break-even process. Economic projections clearly showed that the machine learning regression can differentiate between economically viable and non-viable feeds and, due to the relationship between biocrude yield and economic performance, can even differentiate between economically viable feeds with discernment. When combined with other parameters, including especially process scale and feedstock cost, the new machine learned regression is a valuable tool for guiding investment of finite resources.