Smart sensors, IoT connectivity, and predictive Machine Learning based Digital Twins can transform manufacturing and healthcare systems by enabling real-time data collection, analysis, and intelligent decision-making. For that purpose, training these ML algorithms on synthetic data generated from Discrete Event Simulation (DES) models is a highly promising solution. This paper presents two case studies in manufacturing and healthcare systems, demonstrating the capability of this technique to develop powerful ML surrogate models that can generate efficient estimates in near real-time. Results indicate that advanced preprocessing techniques are essential, particularly when the data exhibit strong skewness or multimodal behavior. The findings highlight that applying common data cleaning and transformation techniques to DES-generated synthetic data, already cleaned during input modeling, may hide valid variations crucial for preserving the dataset's integrity. These results emphasize the need for further research in designing effective preprocessing pipelines that preserve the structure and variability of DES-generated synthetic data.