17β-Hydroxysteroid Dehydrogenase (HSD17β) regulates the formation of androgens and all estrogens. 15 HSD17β isoenzymes have been discovered, which vary in function and tissue distribution. HSD17β4 promotes the production of estrone through the oxidation of estradiol and fatty acids. When metabolized, estrone becomes either non-carcinogenic 2-hydroxyestrone or carcinogenic 4-hydroxyesterone. The HSD17β4 gene has been shown to be upregulated nearly 5-fold in prostate cancer tissues, with no change in HSD17β1, 2, or 3. Additionally, prostate cancer tissues display increased levels of HSD17β4 mRNA as well as increased enzymatic activity, making HSD17β4 a potential target for prostate cancer. The aim of this study was to discover new HSD17β4 inhibitors using QSAR machine learning models. Three NCATS compound libraries (MLSMR, NPC, MIPE) had previously been tested in an HSD17β4 inhibition assay, and were aggregated into one large dataset, yielding a curated modeling set of 75,554 compounds, with 286 compounds showing proficient inhibitory activity (AC50 < 10 μM). This dataset was used to build 22 different QSAR models using various model architectures and feature sets. The best three models (Tree-based (RF and XGBoost) and deep learning-based (DLCA)), as selected by PPV and AUC values based on a 5-fold cross validation procedure, were used for a virtual screening campaign performed on two larger compound libraries. 64 compounds from each library were selected by the three models independently without overlapping. Thus, 348 compounds were nominated and tested in corresponding in vitro assay. This resulted in a 38% hit rate, a 3-fold increase compared to original qHTS hit rate (14%). When looking for more specific activity levels: 14% of the compounds had an an AC50 < 30 μM, and 2% of the compounds had an AC50 < 10 μM. These 9 compounds with an AC50 < 10 μM serve as promising starting points for further medicinal chemistry optimization.