On Development of Machine Learning Models with Aim of Medical Differential Diagnostics of the Comorbid States

Abstract

The purpose of the work is to develop mathematical and software background for the development of machine learning (ML) models in differential diagnostics of comorbid states. Flowchart includes basic steps of ML model development, including the statement of task, the choice of method (learner), setting its parameters and model assessment. The problems dealing with dimension reduction which arise often in differential diagnostics of comorbid states are highlighted and solved with help of modified PCA method. As an example we consider the problem of development of classifier for chronic pancreatitis combined with ascaridosis where we solve all tasks of ML model development. With help of benchmark of learners in the package mlr we compare different methods of ML when applying them in differential diagnostics of comorbid states.