Enabling Pro-active Problem Management by Predictive Modelling : Data mining and statistical analysis of past problems to enable pro-active actions for Internet Protocol Television
Abstract: Service providers strive to guarantee a certain level of quality of their services, to stay relevant and to keep their customers satisfied and to avoid customer churn. IPTV is such a service. To be able to guarantee quality of service and uptime a good problem management system is vital. Problem management system, are a system of handling occurrences and solutions of all problems and errors. Its primary goal is to detect, prevent and solve errors and incidents, but also prevent reoccurring errors and minimizing the time line of existing problems by finding its root cause. On first contact of a problem there is human operators classifying the problem. Problems with problem management is that it is reactive and problems keep happening. To be able to become pro-active, problems need to be predicted before they occur. This thesis evaluates past problems to enable pro-active actions, by statistics and actionable data mining, using performance data from IPTV service and set top boxes, not utilized today by the company this thesis is carried out for. The result shows that pro-active actions are enabled by adding supervision of the performance data and a packet error ratio. The result also shows that some large problems gets classified too late. By utilizing machine learning and predictive modelling: Logistic regression and Artificial neural network, future disturbance incidents can be predicted. Using disturbance data gave a model accuracy of 80% and 80% respectively with a model Matthew correlation coefficient of 0.50 and 0.56 respectively. Given information of affected active service consumers, if provided, customer impact could easier be assessed. The larger part of this project went to the business understanding, data understanding, data preparation and analysis. By utilizing the unutilized performance data, the problem management system can become more proactive, with the addition of supervision of the IPTV performance data and predictive modelling.
AT THIS PAGE YOU CAN DOWNLOAD THE WHOLE ESSAY. (follow the link to the next page)