Problems of training set’s formation in machine learning tasks

Proper formation of the training set is often crucial in the problems of machine learning, that is recognized by most experts in machine learning, often solving machine learning problems is reduced to the competent formation of the training set. Despite this, in the modern literature on machine learning these issues given undeservedly little attention, although often it is the correct formation of the training set is crucial for solving practical problems, theoretical basis practically absent. This article is intended to correct this shortcoming. The article examines the potential problems and errors in the formation of a training set, summarizes the author’s experience in solving machine learning tasks, offers a models for describing the phenomena, associated with the formation of a training set, methods for improving the training set are given. Practical recommendations, based on these theoretical models, are given. At the end of the article shows the experimental results demonstrating some of the problems of training set formation and methods for their solution by the example of learning a decision trees.