	US 12,289,195 B2
	Automated root cause analysis of network issues in a cellular network using machine learning
Michal Horemuz, Danderyd (SE); Jared Keown, Sundbyberg (SE); Rahul Jain, Bangalore (IN); Cesar Marco Rubio, Madrid (ES); Kin Soon Douglas Ho, Mount Waverley (AU); and Philipp Frank, Madrid (ES)
Assigned to TELEFONAKTIEBOLAGET LM ERICSSON (PUBL), Stockholm (SE)
Appl. No. 18/548,715
Filed by Telefonaktiebolaget LM Ericsson (publ), Stockholm (SE)
PCT Filed Mar. 2, 2022, PCT No. PCT/EP2022/055251 § 371(c)(1), (2) Date Sep. 1, 2023, PCT Pub. No. WO2022/184762, PCT Pub. Date Sep. 9, 2022.
Claims priority of application No. 202141008813 (IN), filed on Mar. 2, 2021.
Prior Publication US 2024/0171452 A1, May 23, 2024
Int. Cl. G06F 15/173 (2006.01); H04L 41/0631 (2022.01); H04L 41/14 (2022.01); H04W 28/08 (2023.01)

CPC H04L 41/0645 (2013.01) [H04L 41/145 (2013.01); H04W 28/0958 (2020.05)]

20 Claims

1. A computer-implemented method for analyzing issues in a cellular network, wherein the cellular network includes a plurality of cells and wherein the plurality of cells include a plurality of source cells and a plurality of neighbor cells, the method comprising:

obtaining data for each cell of the plurality of cells;

building a network graph, using the data obtained, representing features of the plurality of cells, wherein, for each source cell of the plurality of source cells, each neighbor cell of said each source cell is ranked based on the data obtained for said each neighbor cell;

identifying, using the network graph, a plurality of sub-graphs for said each source cell indicating a network issue, wherein each sub-graph of the plurality of sub-graphs represents features of said each source cell and a set of neighbor cells selected based on rank and features of the plurality of neighbor cells;

for each network issue of said each source cell for said each sub-graph of the plurality of sub-graphs, ranking each feature of said each neighbor cell represented in said each sub-graph for said each source cell and identifying a set of ranked neighbor features;

for each network issue of said each source cell for said each sub-graph of the plurality of sub-graphs, ranking each feature of said each source cell for said each sub-graph and identifying a set of ranked source features;

identifying, using the set of ranked neighbor features and the set of ranked source features, a feature set for all sub-graphs, wherein the feature set includes all or a reduced set of features;

for said each source cell, training, using the feature set identified, a first machine learning (ML) model for said each neighbor cell in the set of neighbor cells for said each source cell and determining a neighbor contribution value for said each neighbor cell, representing the contribution of said each neighbor cell to said each network issue;

for said each source cell, training, using the feature set identified, the first ML model for said each source cell and determining a source contribution value for said each source cell representing the contribution of said each source cell to said each network issue;

for said each network issue for said each source cell, using the neighbor contribution values and source contribution values to identify a set of contributing features from the feature set identified that contribute to said each network issue;

building, using the set of contributing features, a clustering model having a plurality of clusters, wherein each cluster of the plurality of clusters identifies a pattern of one or more neighbor cells contribution to said each network issue;

training a second ML model, using the patterns identified by each cluster of the plurality of clusters, to classify an unidentified pattern of one or more neighbor cells contribution to said each network issue and identify root cause information for said each network issue.