Virtual Patient Model: An Approach for Generating Synthetic Healthcare Time Series Data

In this paper, we address the problem of research data availability and access in the healthcare sector, by proposing Virtual Patient Model (VPM) a process that combines the use of an optimization algorithm, statistical analysis and machine learning techniques to generate synthetic time series data and report their effectiveness in predictive models. We validate the proposed model by implementing a genetic algorithm that captures important features of a real-world patient time-series data (original) and in applying constraints in the generative process, outputs a best fit synthetic candidate solution of the original time series data. Experimental results using statistical verification tests on both the synthetic and original datasets showed that the synthetic dataset preserved features from the original dataset. We used machine learning prediction models that integrated the synthetic dataset into classification learners and compare their outcomes with those from learners trained with the original dataset. We found promising results in machine learner's ability to discriminate between the different classes when synthetic data is used for training.