O-RAN Xapps Conflict Prediction Using Graph Convolutional Networks

O-RAN hosts many intelligent applications known as eXtended Applications (xApps). xApps are applications that leverage advanced algorithms to make dynamic decisions for network optimization. Each application operates with distinct optimization objectives and is managed by independent operators while accessing shared network resources. Conflicts occur when a deployed xApp's objective interferes with another xApp, resulting in incompatible actions or decisions that may negatively impact network performance. The lack of a unified mechanism to coordinate and prioritize actions across different applications can lead to various types of conflicts. Conflict prediction in O-RAN is the proactive analytical process by which potential interactions or behaviors that may lead to conflicts between network applications are identified in advance, before they manifest in the operational system. In our paper, we introduce a novel data-driven Graph Convolutional Network-based method called GRAPH-based Intelligent xApp Conflict Prediction and Analysis (GRAPHICA). It predicts three types of conflicts and pinpoints the root causes. GRAPHICA captures the complex, hidden dependencies among the xApps, controlled parameters, and key performance indicators in O-RAN to predict potential conflicts. Then, it identifies the root causes contributing to the predicted conflicts. The proposed method is evaluated using highly imbalanced synthetic datasets, in which conflict instances constitute between 40 % and merely 10 % of the data. This evaluation setting is designed to reflect realistic operational environments where conflicts are infrequent, thereby enabling a comprehensive assessment of the model's performance under real-world conditions. Experimental results demonstrate an F1-score over 98 % for the synthesized datasets with varying levels of class imbalance.