How to Generate Best-in-Class Synthetic Time Series Data

This post discusses the process of generating high-quality synthetic time series data using Gretel's DGAN and Gretel Tuner. The goal is to create data that not only statistically resembles the original data but also maintains logical consistency and sequence of events. The dataset used in this example focuses on project management lifecycle events, including initiation, planning, execution, monitoring and controlling, and closure. Mandatory and optional events are defined within these stages. To generate synthetic time series data, Gretel's DGAN model is fine-tuned using the Gretel Tuner to optimize a custom metric that captures the statistical properties of the original dataset. The metric considers the distribution of event types and probabilities of event transitions in both the original and synthetic datasets. The time series data is prepared for DGAN model training by identifying the maximum sequence length across the dataset, ensuring all sequences are of equal length through padding. The Gretel Tuner config is then defined to optimize the DGAN model settings. Once the optimal DGAN model settings are found using the Gretel Tuner, synthetic time-series data can be generated and validated by comparing it with the original dataset. This approach enables generating data compliant with intricate business rules, providing a robust solution for simulations, testing, and enhancing data privacy.