Supervisors: Prof. Orçun Göksel
Medical triage is the process of determining the priority of patients' treatments based on the severity of their symptoms. Currently a care provider and treatment urgency suggestion are given by doctors and triage nurses, but it is desirable to automate and normalize this process using data-driven approaches. In this thesis, firstly a suggestion confidence score is proposed for predicting posterior probabilities of triage outcomes based on probability calibration methods and performance-rejection rates of several heuristics used in machine learning. Such confidence heuristics based on the calibrated probabilities of the classifier achieve a fairly good trade-off between performance and rejection, while the problem of unbalanced class performance of both classifier and confidence metrics has been mitigated by probability calibration. Secondly, an interactive Q&A system has been developed based on pseudo-relevance feedback-based query expansion techniques in order to imitate triage doctors questions to the patient regarding additional relevant symptoms in order to arrive a correct decision.