КЛАСИФІКАЦІЯ ВРАЗЛИВОСТЕЙ ПРОГРАМНОГО ЗАБЕЗПЕЧЕННЯ ТА ФОРМУВАННЯ ОЗНАК ЇХ НАЯВНОСТІ НА ОСНОВІ МЕТОДІВ МАШИННОГО НАВЧАННЯ

The object of the study is the process of vulnerability detection in software. The subject of the study includes approaches to vulnerability classification and methods for forming indicators of their presence suitable for machine learning algorithms. The aim of the work is to systematize existing types of security defects and to develop a method for feature extraction and preprocessing that enables their formal representation within a numerical feature space. The paper analyzes the main principles of vulnerability classification according to root causes, exploitation techniques, manifestation levels, and impact on security properties. A generalized system of structural, behavioral, and contextual parameters that may serve as indicators of potential defects is proposed. Procedures for transforming heterogeneous data are developed, including normalization of numerical values, encoding of categorical attributes, binarization of logical features, and dimensionality reduction. The obtained results provide a methodological basis for the further development of intelligent systems for automated software vulnerability detection.