Identification of IT Incidents for Improved Risk Analysis by Using Machine Learning

Today almost every system or service, e.g., water, power supply, transportation, etc. is dependent on IT systems, and failure of these systems have serious and negative effects on the society. IT incidents are critical for the society as they can stop the function of critical systems and services. Moreover, in a software engineering context risk analysis is an important activity for the development and operation of safe software-intensive systems. However, the increased complexity and size of software intensive systems put additional requirements on the effectiveness of the risk analysis process. Therefore, the risk analysis process needs to be improved and it is believed that by having an overview of already occurred IT incidents, the risk analysis process can be improved. The saved information about IT incidents can be used as an input to risk analysis, which can help to correctly estimate the consequences of potential risks. This study investigates how difficult is it to find relevant risks from the available sources and the effort required to set up such a system. It also investigates how accurate are the found risks. It presents a prototype solution of a system that automatically identifies information pertaining
to IT incidents, from texts available online on Internet news sources, that have happened. This way IT incidents can be saved semi-automatically in a database and the saved information can be used later as an input to risk analysis. In this study 58% of texts that potentially can contain information about IT incidents were correctly identified from an experiment dataset by using the presented method. It is concluded that the identifying texts about IT incidents with automated methods like the one presented in this study is possible, but it requires some effort to set up.