Additive Manufacturing (AM) process parameters significantly influence the thermal, mechanical, and surface roughness properties of printed parts. While experimental characterization is highly accurate, it is often excessively time-consuming and costly, hindering the development of scalable predictive models. This study adopts a Materials Informatics approach to curate a multi-sourced benchmarking dataset comprising 543 entries for ABS, PLA, and TPU filaments, integrating data from web-scraping (WSDS), literature (LDDS), and manufacturer technical datasheets (MTDS). Data cleaning included outlier detection via Tukey’s fences and the evaluation of imputation strategies, with K-Nearest Neighbors (KNN) consistently outperforming simple mean/median methods. We evaluated three regression models, Random Forest (RF), Ridge Regression (RR), and Support Vector Regression (SVR), using a nested cross-validation pipeline to predict tensile strength, elongation, and surface roughness. The results reveal significant data gaps in existing literature, with missingness exceeding 60% for critical parameters like fan speed and wall thickness. While the Random Forest model achieved high predictive accuracy for bulk mechanical properties, specifically a test-set R2 of 0.995 for PLA elongation, all models struggled with surface roughness, often yielding negative R2 values. This suggests that current process descriptors are insufficient to capture the stochastic nature of surface morphology. This work provides a foundational benchmarking dataset for the research community and highlights the necessity for stricter data mining protocols and more sophisticated feature engineering to achieve reliable predictive quality in material extrusion (MEX) polymers.