Method, system and computer program for anomaly detection in mobile cellular networks

6:

• The method of claim 1, wherein the prediction of the expected KPI values in the target cell over the predicting sliding window further comprises:
• associating weights to the connected neighboring cells based on the past time sliding window;
• and
• making the predictions based on the associated weights and considering an evolution of the KPIs in the connected neighboring cells during the prediction sliding window.

8:

• The method of claim 1, wherein the calculation of the anomaly score further comprises using:
• error metrics between the predicted expected KPI values and the actual measured values;
• or
• a dynamic score function that considers accumulated errors from previous calculations.

14:

• A system for anomaly detection in cellular networks, comprising a memory and at least one processor, wherein the at least one processor is configured to:
• generate a graph representation of a plurality of cells of a mobile cellular network, where each cell of the plurality of cells represents a node in the graph representation and nodes are connected by edges according to a predefined criteria;
• select one of the cells of the graph as a target cell;
• obtain monitored key performance indicators, KPIs, of connected neighboring cells of the target cell got over a past time sliding window and a prediction sliding window and obtain monitored KPIs of the target cell got over the past time sliding window;
• predict expected KPI values in the target cell over the predicting sliding window using the generated graph representation and the obtained monitored KPIs of the connected neighboring cells and of the target cell by a machine learning, ML, context-based predictor;
• make context-based predictions on the monitored KPIs of the target cell by training the ML context-based predictor using the predicted expected KPI values in the target cell over the predicting sliding window;
• calculate an anomaly score by comparing the predicted expected KPI values in the target cell over the predicting sliding window with actual values measured in the target cell over the same period;
• and
• determine that an anomaly occurs when the calculated anomaly score exceeds a predefined detection threshold.