Analysis and fault detection in internet traffic data with time series analysis and NLP approches.

Telecommunication companies have several transmission systems scattered over the territories that use Internet services. These transmission systems are subject to malfunctions that must be recognized in order to be resolved as quickly as possible. The types of failures can be multiple as transmissions occur on different frequencies, directions, technologies and so on, so it is complicated to recognize the actual presence of a failure and its type. In this study, we intend to approach the problem through machine learning algorithms that aim at automatic recognition of failures and their types. This study intends to investigate the problem through the use of time-series analysis techniques related to data traffic volume data and NLP for reading customer complaint tickets during the same periods. An important goal of this work is to be able to distinguish between genuine failures in transmission networks from physiological drops in data volume (e.g., periodic drops). The ultimate goal is to obtain faster and more correct reports so that resources can be allocated for resolution in the shortest possible time and in an automatic manner.