A Mixture Model Diagnosis System

Diagnosis is the process of identifying the disorders of a machine or a patient by considering its history, symptoms and other signs. Starting from possible initial information, new information is requested in a sequential manner and the diagnosis is made more precise. It is thus a missing data problem since not everything is known. We model the joint probability distribution of the data from a case database with mixture models. Model parameters are estimated by the EM algorithm which gives the additional beneet that missing data in the database itself can also be handled correctly. Request of new information to reene the diagnosis is performed using the maximum utility principle from decision theory. Since the system is based on machine learning it is domain independent. An example using a heart disease database is presented.