Detecting Cell Outage by Applying Density Based Anomaly Detection Algorithm Using Machine Learning Technique: The Case of Ethio-Telecom UMTS Network
No Thumbnail Available
Date
2020-02-01
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Addis Ababa University
Abstract
This thesis develops a model to detect cell outage on real ethio telecom network data by using
density based anomaly detection algorithms through the application of machine learning
technique. Cell outage is the total/partial loss of radio network in a given area and the process of
detecting cell outage is called cell outage detection. Cross-Industry Standard Process (CRISPDM)
machine learning methodology, which is a six staged open standard process model that
describes common approaches for data mining or machine learning was used. The considered
data was normal and problematic network environment obtained from ethio telecom UMTS
network in Addis Ababa. The proposed detection framework used network performance data
(incoming handover (inHO), which is the process of transferring an ongoing call from one cell to
the other, and traffic data, originated from base station and terminated to mobile device) of the
neighbor cells to capture the normal network state and to detect the outage of the target cell
automatically in a pre-set time interval in the UMTS network environment.
To profile the normal network operation, the study used two density based anomaly detection
algorithms; namely, the K- Nearest Neighbor (K-NN) and Local Outlier Factor (LOF) algorithms,
of which one was selected based on its performance during the training. To validate the models,
K-fold cross validation technique was used and for the selection of the optimal model, parameter
selection was done for different values of K (K=1,2, 3…30). To compare the two algorithms,
Receiver Operating Characteristic (ROC) curves were used. Based on the results, the K-NNAD
was found to be of a better performance than the LOFAD, thus was selected as a detector in the
profiling stage. The proof of the system model was tested by using real problematic network state
data and the results of classic data mining metrics were obtained. Based on the results obtained
from the testing, the K-NNAD method was found to perform better in detecting outage cells in the
proposed framework.
Description
Keywords
Cell outage, Machine learning, Anomaly detection, Self-healing, UMTS network, CRISP-DM