Sequence-Walking Decision Tree for Multivariate Healthcare Data

Patients, whether they are currently hospitalized or cured at home via telemedicine, are constantly monitored by stationary/wearable devices that provide accurate, yet an overwhelming amount of temporal data. Due to the heterogeneous nature of such data, reconciling them in order to support decisions about the patient in an effective, interpretable, and explainable way is not a trivial task. In this paper, we propose a framework for turning a large collection of heterogeneous intra-patient data into a uniform representation based on event-logs; moreover, on such a representation, we introduce a new tree-based decision model for performing tasks such as classification, early detection, data mining, and cluster-based similarity.