The Artificial Intelligence Based Diagnostic Assistant – AIDA

Even though interest in Machine Learning Based Clinical Decision Support Systems (ML-CDSS) has been rapidly growing in recent years, most research and development is exclusively focused on secondary and tertiary care – even though effective diagnostic support in primary care could significantly improve both the circumstances, the process and the outcome of general practice. In this paper, we study the suitability of five supervised machine learning algorithms to the problem of multiclass classification with sparse Boolean features on a primary care data set, and we examine the robustness of the algorithms to incomplete data. We introduce our own classification algorithm, the Artificial Intelligence Based Diagnostic Assistant (AIDA), which is capable of incorporating both symptoms and contextual information into its diagnostic process, thus modeling the decision-making of physicians in a novel and accurate manner. Through our experimental results we show that AIDA is by far the most suitable classification algorithm for ML-CDSS applications in primary care, owing to its high accuracy and outstanding robustness to missing, sparse information.