Causal relationships in classification problems

In the present paper, a classification object is considered as the cause for the appearance of one or more consequences and any classification algorithm decides on the class observing the consequences from the analyzed cause. The paper considers the consequences of the cause in the binary classification problem as sources of additional information confirming or rejecting the hypothesis of the cause in the classified object. When considering a hypothesis about the presence or absence of a certain cause in an object classified by this property, the knowledge presentation language is automatically built based on several consequences. Then, it is easy to use the available information from different information spaces in an object classification task. To use cause-and-effect relationships in a classification task, machine learning should be used. In conditions of teaching with a teacher, there are many precedents when the presence of a cause is known. Then one can statistically single out events that are the consequences of the cause. Deterministic cause-and-effect relationships generate errors only at the expense of noise. In those precedents where there is no cause, positive classification appears only at the expense of noise regardless of precedent to precedent. Thus, even a weak deviation from equally probable noise allows one to build a consistent criterion that distinguishes consequences from random noise. Sequelae can be isolated independently of each other. This follows from the determinism of the cause-and-effect relationship and the independence of noise.