Modèles probabilistes et statistiques/d'apprentissage automatique avancés pour la détection d'anomalies : application dans le secteur des télécommunications
Michel Kamel · HAL (Le Centre pour la Communication Scientifique Directe) · 2023
The exponential growth of connected device networks worldwide means that telecommunications operators need intelligent and efficient systems to help maintain their vast and complex networks. To address the limitations of the most popular anomaly detection (AD) models, the authors propose a new multidimensional probabilistic geometric model to search for abnormal behaviors in the data space, generate anomaly scores, and quantify anomaly factors. They also introduce an algorithm to generate a final score based on four features derived from historical data for alarm data. Additionally, they present an algorithm to assist in preprocessing textual data, clustering them into classes, and dynamically labeling each class as an anomaly or not. Finally, they propose a method that reduces dimensionality and provides an anomaly score system based on the theory of records. Overall, their research provides innovative methods for detecting and prioritizing anomalies in telecommunications networks and provides powerful tools for data analysis and network maintenance.