Predikce událostí v metrických datech za pomoci analýzy časových řad

The goal of the master’s thesis is to develop a module for an open source monitoring and management platform Hawkular. The module should provide real time predictive capabilities for collected time series data. Proposed solution should be computationally efficient and more importantly, it should autonomously select the best time series model for given metric. In other words it should work without analyst ’s interaction. The master’s thesis starts with time series theory and analysis of various approaches for time series forecasting. It discusses which models are best for Hawkular requirements. The implementation chapter focuses on the most important parts like time series models, automatic forecasters and integration into Hawkular. The last chapter conducts evaluation of the implemented models. Models are benchmarked against its alternatives from statistical language R. The results show that implemented models are similar to R alternatives from the prediction accuracy perspective.

University

Faculty of Informatics

Date of Completion

spring 2016

Resources

Leader

Adam Rambousek

Consultant

Jiří Kremser

Student

Pavol Loffay