CLUSTERING SOFTWARE DEVELOPERS AS A TOOL FOR PREDICTING SOFTWARE CODE DEFECTS

The article is devoted to the development of a methodology for clustering software developers to predict the probability of errors in code using artificial intelligence. The paper analyzes the criteria for grouping developers, including programming style, dynamic activity metrics, error history, cognitive features and technological expertise. A system for assessing the quality of clustering through compactness metrics and external indicators of correlation with the frequency of defects is proposed. The use of machine learning algorithms (CNN, LSTM, GNN) and tools (Scikit-learn, TensorFlow, CodeBERT) allows you to automate data analysis and generate personalized risk forecasts. Experiments demonstrate an increase in forecasting accuracy due to taking into account interpersonal variability, which opens up opportunities for optimizing testing processes, team training and implementing adaptive quality control systems.