The development and application of methods of preliminary processing of tabular data for solving problems of multivalued classification of computer attacks is considered. The object of the study is a data set containing multivalued records collected using a hardware and software complex developed by the authors. The analysis of the attributes of the dataset was carried out, during which 28 attributes were identified that are of the greatest informational importance when used for classification by machine learning algorithms. The expediency of using autoencoders in the field of information security, in tasks related to datasets with the property of ambiguity of target attributes is substantiated. Practical significance: data preprocessing can be used to improve the accuracy of detecting and classifying multi-valued computer attacks.

Keywords: information security, computer attacks, multi-label, multi-label classification, multivalued classification, dataset analysis, experimental data collection, multivalued data, network attacks, information security