As the term “synthetic” suggests, synthetic datasets are generated through computer programs, instead of being composed through the documentation of real-world events. The primary purpose of a synthetic dataset is to be versatile and robust enough to be useful for the training of machine learning models.