The choice of method of binary classification with technical diagnosis using machine learning

For carrying out technical diagnostics can be used various methods of machine learning. The main task of binary classification in relation to the diagnosis of technical objects is determined by the specified parameters for the functioning of the object, whether it is healthy. It is assumed that there are many precedents: situations with the specified parameters of the functioning and the famous State of the object. The task of separating objects into two classes-serviceable and unserviceable may be solved as using classical statistical methods, for example, discriminant analysis, and using modern computer technologies based on machine learning methods. You can try to build compositions of various algorithms. Experience shows that two of the main compositional method-bjegging and busting-give much more accurate results than using a separate algorithm on a specific set of data. The quality of binary classification (healthy or unhealthy state of the object) is estimated by various criteria: percentage of errors in the control sample, F-measure and the criterion of AUC-ROC-area under the curve of errors. The problem of selecting the best method of classification by the specified criteria.