Despite self-organizing networks (SONs) pursue the
automation of management tasks in current cellular networks,
the selection of the most useful performance indicators (PIs),
used as inputs for SON functions, is still performed by network
experts. In this letter, a novel supervised technique for the
automatic selection of PIs for self-healing functions is proposed,
relying on the dissimilarity of their statistical behavior under
different network states. Results using data from a live network
show that the proposed method outperforms an expert’s
selection, allowing the volume and complexity of both network
databases and SON functions to be reduced without an expert’s
intervention.
