A personalized committee classification

应用背景 A Personalized committee classification algorithm To select training samples, we created a patient–patient network based on co-expression, and used a well-known graph algorithm to identify patients that are topologically close to the representing patient in the network. Each base classifier was then  evaluated for its prediction accuracy on its representing patient and was discarded if the prediction was incorrect. This procedure was repeated for all patients in the training dataset, resulting in a rich pool of classifiers that could be used for predictions in the test dataset. New patients were classified by a subset of these base classifiers chosen based on the similarity between the target patient and the representing patients of the base classifiers, and predictions from individual base classifiers were combined to form the final decision.