Using Synthetic Data for Improving Robustness and Resilience in ML-Based Smart Services

We set to answer the question of whether robustness and resilience of machine learning (ML) based smart services in the Internet-of-Things (IoT) context can be improved by using synthetic data. These data can be in the form of training data for ML algorithms or service interactions. While there is plenty of research on the use of synthetic data in general ML models, there is a lack of understanding on the use of synthetic data in the smart service context. This can help make smart services more resilient by solving the cold-start problem and improve their generalization capabilities. We propose an architecture for ML-based smart services that integrates both real and synthetic data and perform an empirical evaluation than combines publicly available sensor data (streamflow data) and state-of-the-art synthetic data generation methods. Using standard performance metrics, our results show that enhancing a dataset with synthetic data can improve performance significantly even with a modest amount of data.