Anomaly Detection in LTE Mobile Network / Mahmoud Adel Nour

Supervisor:
Nashwa AbdelBaki

Thesis (M.A.)—Nile University, Egypt, 2022 .

"Includes bibliographical references"

Contents:
CERTIFICATION OF APPROVAL ...................................................................... iii
DECLARATION ................................................................................................... vii
ACKNOWLEDGEMENTS .................................................................................... ix
TABLE OF CONTENTS ........................................................................................ xi
LIST OF FIGURES .............................................................................................. xiii
LIST OF TABLES ................................................................................................. xv
ACRONYMS ....................................................................................................... xvii
PUBLICATIONS .................................................................................................. xix
ABSTRACT .......................................................................................................... xxi
1. CHAPTER ONE HISTORICAL BACKGROUND ......................................... 1
1.1. Motivation ................................................................................................. 1
1.2. Background and Aims ............................................................................... 2
1.3. Thesis Structure and Contribution............................................................. 4
2. CHAPTER TWO STATE OF THE ART ......................................................... 7
2.1. Main Streams............................................................................................. 8
2.1.1. Statistical Modelling .................................................................................... 8
2.1.2. Regression Scoring....................................................................................... 9
2.1.3. Supervised Learning .................................................................................. 10
2.1.4. Unsupervised Learning .............................................................................. 10
2.1.5. Autoencoders ............................................................................................ 11
2.1.6. Policy-based Reinforcement Learning ....................................................... 12
xii
2.2. Summary and Conclusion ....................................................................... 12
3. CHAPTER THREE EXPLORATION AND EXPLOITATION OF THE LTE MOBILE NETWORK DATA ............................................................................... 15
3.1. Problem Definition .................................................................................. 16
3.2. Data Description ...................................................................................... 16
3.3. Dataset Exploration ................................................................................. 19
3.4. Summary and Conclusion ....................................................................... 28
4. CHAPTER FOUR EXPERIMENTAL PROCESS AND RESULTS ............ 31
4.1. Experiment Architecture ......................................................................... 32
4.2. Results ..................................................................................................... 61
4.3. Discussion and limitations ...................................................................... 64
5. CHAPTER FIVE CONCLUSION AND FUTURE WORK .......................... 67
5.1. Future Work ............................................................................................ 68
References .............................................................................................................. 71

Abstract:
The usage of anomaly detection is widely known for its benefits. It is defined as the outlier detection mechanism. The techniques added to anomaly detection is huge since the 2000s. The start of the anomaly detection was to classify the cases whether it is normal cases or anomaly cases. Over time, the time series challenges need to engage with the anomaly techniques. The time series challenges were handled with the statistical approaches at the beginning. Then the machine learning techniques added more information to the assessment. Supervised learning is used to classify the anomaly cases versus the normal cases using the available data. Then the unsupervised learning is injected to profile the anomaly cases. The unsupervised learning showed enhancement in the cases where the anomaly cases were clear enough to classify.
The combination of the multivariate appeared to enhance the results with the increase in the business representation. The increase in the business representation gives the flexibility of defining multiple domains. These domains could represent the business view. We use the power of unsupervised learning enhanced by statistical modeling to solve the anomaly detection challenge.
In this thesis, we propose a framework for detecting abnormal EUtranCell in LTE networks using raw counters of LTE mobile networks. The LTE mobile network is known for its huge number of cells in the network. The number of profiles in the LTE mobile network depends on how old the mobile operator is. Our method is based on using unsupervised learning to model health network KPIs in terms of revenue, performance, and customer satisfaction. This modeling gives the edge of getting the fit EUtranCell profile. We also use statistical modeling to set the severity of cases to guide mobile network operators to healthy networks. This provides flexibility in terms of operating costs relative to customer perception requirements. Our framework gives the flexibility of using any approach to assess anomaly cases.

Text in English, abstracts in English.