Efficiency Model Selection in AutoML for CVD Risk Prediction

Cardiovascular disease is the leading cause of mortality not only in Australia but also worldwide. There is a high demand for a machine learning system that can predict disease risk at an early stage so that healthcare experts can provide interventions. The process of building a machine learning system requires cooperation between machine learning experts and healthcare experts who understand the data. This is both labor and computation-intensive. To make the state-of-the-art machine learning models accessible and save computation costs, we introduce a training data size prediction model called EDSPA, which is build for an automated machine learning system called Auto-Imblearn. EDSPA can predict the minimal number of training instances required for Auto-Imblearn to determine an ideal pipeline. Compared with the state-of-the-art automated machine learning systems, our system achieves the best prediction accuracy whilst saving computation costs when compared with exhaustive search.