Reliable diagnosis of acute abdominal pain with conformal prediction
Harris Papadopoulos, Alex Gammerman, Vladimir Vovk · International journal of engineering intelligent systems for electrical engineering and communications · 2009
Medical decision support is an area in which a lot of machine learning research has been conducted and several diagnostic and prognostic systems have been developed. The majority of these systems only produce bare predictions, without any indication of how reliable each of these predictions is. An indication of this kind however, is highly desirable especially in the medical field. In this paper we address this problem with the use of a recently developed technique, called conformal prediction, for accompanying the predictions of traditional machine learning algorithms with measures of their accuracy and reliability. We apply conformal prediction based on a Neural Network classifier to the problem of acute abdominal pain diagnosis and obtain predictions which have a high level of accuracy and are complemented with well-calibrated and practically useful confidence measures.