Comparative Analysis of Knowledge Base System Development Approaches in the Context of Ethio Telecom
No Thumbnail Available
Date
2022-01
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Addis Ababa University
Abstract
For a successful business, telecom companies need to provide consistent, quality
and timely maintenance and support to their customers. This is difficult to
achieve without timely availability of domain experts that can efficiently be managed
using the Knowledge Base System (KBS). Different approaches are applied
to implement KBS and researchers study accuracy performance of the approaches.
However, there are limited performance studies on the approaches in the context
of telecom companies and none for ethio telecom. Most of the studies are also
only focused to accuracy performance of the approaches and other relevant performance
metrics that have an impact on the incident handling time of Information
Technology (IT) experts are not addressed well.
This thesis work presents performance comparison for KBS development approaches
in the context of operation support and maintenance process of ethio telecom.
Rule-based, Case-based and Artificial Neural Network (ANN)-based KBS development
approaches are considered and these approaches are modeled using data
collected from ethio telecom technical support experts’ knowledge and experience
through interviews and data analysis. For the performance analysis, data preprocessing
techniques including data reduction, data cleaning, data integration and
data transformation are applied on the collected datasets. Execution time, accuracy,
F-measure, precision, recall, error rate and mean square error metrics that
have an impact on incident handling time are used for the performance evaluation.
Python programming language is used to model, train, evaluate and compare
the selected approaches.
Achieved performance results show that ANN-based KBS development approach
has a better performance in terms of building and executing the model, taking a
minimum time of 0.95 seconds. Furthermore, rule-based approach accomplishes
the highest accuracy that is 98.6 %, with minimum error in identifying the class
labels which are incident types. Based on the evaluation results on the execution time or response time, accuracy, and error measures which have an impact on the
incident handling time, a hybrid KBS development approach from the ANN and
rule-based KBS development approach can provide the best aggregate result.
Description
Keywords
Rule-based development approach, Case-based development approach, ANN based development approach