METHODOLOGY OF INTELLECTUAL ASSESSMENT OF INFORMATION SECURITY RISKS OF CORPORATE DATABASES

Abstract

The article proposes a methodology for intelligent risk assessment of corporate database security based on the use of an ensemble approach to anomaly detection. Based on the analysis of modern approaches and research in database security, it is concluded that the use of artificial intelligence in protection systems is necessary to account for the complex impact of negative factors on the security of corporate databases. Unlike traditional methods that focus on individual aspects of user behavior, the proposed solution provides a multidimensional analysis through the integration of multiple machine learning models. In particular, the Isolation Forest algorithm is used to detect point anomalies in the feature space, the Long Short-Term Memory model is applied for analyzing temporal dependencies, and the Autoencoder is utilized to identify structural deviations in multidimensional data. An integrated anomaly score is proposed, which is formed by a weighted combination of the outputs of individual models, enabling improved detection accuracy for complex attack scenarios. Based on the obtained anomaly score, a transition to risk assessment is implemented, taking into account the criticality of resources and the types of database access operations. This approach enables adaptive decision-making for responding to information security incidents. A software prototype of the proposed methodology has been developed using the Python programming language with modern machine learning libraries. An experimental study was conducted on a synthetic dataset simulating both normal and anomalous access scenarios. The obtained results confirm the improved effectiveness of the ensemble model compared to individual approaches in terms of Precision, Recall, F1-score, and ROC-AUC metrics. The proposed methodology can be applied in Security Operations Center (SOC) systems for automated anomaly detection and real-time risk assessment.