Prediction of frequent out-of-hours’ medical use

Abstract

The thesis of this research is that through the mining of Electronic Medical Records containing mixed types of data, and extracting patterns from the processed data, patients can be successfully categorised through means of supervised machine learning early in their engagement with health care providers. This categorisation has quite narrow parameters: the aim of which is identify patients that are less suitable to the health care provider being examined in the course of this research; specifically that of an out-of-hours health care cooperative (OOHC). The motivation for this is to provide potential means for interventionist healthcare, in line with the increasingly role that decentralised regional programmes are having in the avenues of treatment available for patients [19] and the increasing emphasis on community based intervention [24]. The patients in question are frequent users of the OOHC, and represent a small cohort within the dataset as a whole. Our classification methodology, based upon recurrent neural networks, achieves an Area Under the Curve of between 0.81 and 0.92 in the identification of these patients.