在電信服務供應商中，業務支撐系統(business support systems)是個占有舉足輕重的元件，主要用來提供給顧客多種服務，包含產品管理、用戶管理、收入管理等功能。然而，隨著系統複雜度的增加，無可避免的會發生系統故障的問題，這些故障的發生會使的系統無法正常提供服務給客戶，導致電信商有大量的虧損，因此，系統預警變成了一個重要的議題。目前已經有許多研究指出，機器學習在預測問題中是個可行的解決方法。因此，我們的研究想要開發一個以機器學習技術為方法的系統監控平台。我們的平台包含了三個主要模組：資料蒐集、資料分析、與資料視覺化。此外，我們平台也包含了兩大特色，第一，我們開發了完整的監控機制可以蒐集系統中主要元件的效能指標，第二，我們利用多種機器學習的演算法來做系統故障的預測，並且做了一個全面性的比較，找出哪種演算法比較適合使用於電信系統上。

In telecommunication service provider, business support systems are the important components, which offer various business operations toward customers. However, as the complexity of the system is increasing, it is unavoidable that the system may occur defects. Thus, the emergency defects prediction for the systems has become an important issue. Many studies showed that machine learning is a feasible method to improve the prediction problem. Our study aims at developing a monitoring platform for resolving emergency defects problems. Our monitoring platform has three components: data collector, data analyzer, and data visualization. And our platform has two features. First, a comprehensive monitoring mechanism that collects multiple system components’ resource utilization. Second, a friendly user interface for administrators to view the state of system. Besides, we conduct a comparative study of a few of well-known machine learning algorithms. And we evaluate the performance of these algorithms using some standard and widely used performance metrics.

Chapter 1 Introduction 1
Chapter 2 Overview of Machine Learning and Algorithms 4
2.1. Terminologies in machine learning 4
2.2. Machine learning category 5
2.3. Machine learning algorithms 6
Chapter 3 Monitoring system for Business Support System 10
3.1. A BSS deployment architecture 10
3.2. Monitoring system 12
Chapter 4 Implementation of Data Analyzer 17
4.1. Data preprocessing 17
4.2. Feature selection 20
4.3. Machine learning algorithms 21
4.4. Validation 21
4.5. Performance analysis 22
Chapter 5 Experiments and Results 24
5.1. Data preprocessing results 24
5.2. Overview of Weka 25
5.3. Performance of predictive models 25
Chapter 6 Conclusion 28
Bibliography 29

[ 1 ] Docker. [Online]. Available: https://www.docker.com/.
[ 2 ] JBoss Enterprise Application platform. [Online]. Available: https://developers.redhat.com/products/eap/overview/.
[ 3 ] Apache web server. [Online]. Available: https://httpd.apache.org/.
[ 4 ] Apache Module mod_proxy. [Online]. Available: https://httpd.apache.org/docs/2.4/mod/mod_proxy.html.
[ 5 ] Apache Module mod_proxy_balancer. [Online]. Available: https://httpd.apache.org/docs/2.4/mod/mod_proxy_balancer.html.
[ 6 ] Apache JMeter. [Online]. Available: http://jmeter.apache.org/.
[ 7 ] Apache Module mod_status. [Online]. Available: https://httpd.apache.org/docs/2.4/mod/mod_status.html.
[ 8 ] Chart.js. [Online]. Available: http://www.chartjs.org/.
[ 9 ] ARFF format. [Online]. Available: http://www.cs.waikato.ac.nz/ml/weka/arff.html.
[ 10 ] Weka. [Online]. Available: http://www.cs.waikato.ac.nz/ml/weka/.