The monitoring system of Business support system with emergency prediction based on machine learning approach

Abstract

When Business support systems (BSS) suffer poor performance, the system administrator has to spot the problem of BSS as soon as possible. The monitoring system is a great help to quickly gain insight of the system from all sides. By the monitoring system, the administrator can evaluate various metrics on the different fields in one single arranged page and understands the whole picture of current system status from brief reports. Although this kind of problem solving flow does well in certain circumstance, it may still be too late for some critical BSS. To further improve the speed of trouble-shooting, administrator sets the thresholds to each performance metric. Those thresholds are set based on knowledge and experience of administrator. If the performance metric collected from the system is over the set threshold, monitoring system will send alerts to administrator to inform current BSS status, and he can check the system status in advance before the situation get worse. This kind of traditional approach finds the problem about ten minutes before the emergency break out. In our work, we leverage a machine learning approach to determine emergency earlier. The machine learning model we developed can predict the healthy status of the system before the emergency an hour with average 14 points error.