Rolling validation techniques in predition of time series

OData support
Gáspár Csaba
Department of Telecommunications and Media Informatics

Nowadays, it becomes increasingly important to process and analyze the vast amount of data available. We can define different relationships, including what we can predict for the future. In today's world, more and more industries are transformed to analyze their data, as they are even prepared to respond to future events. These include the energy industry. In the energy industry the energy is often used as previously acquired data so that it is possible to estimate how much consumption can be expected in the near future. In that case, they will be prepared for the future when they do not have enough capacity and plan the system fine-tuning in advance to be prepared for that period as well.

My task is also related to this, in which I can help with the analysis of previous data by creating automation, which facilitates the analytical work of users. For analysis, I rely on timeline data, where the difference between the past and the future is easily grasped by the fact that the data is timely interlinked. I will use the rolling evaluation method for modeling models, which does not define a model, but a model building strategy. I implement a Python package that anyone can download through the Internet and use the functions in it to significantly accelerate the time required for data analysis. My job is to devise this automation with its related functions, documenting and publishing functions on the web so that anyone can simply download it and use it immediately. Then I'll show you on real data sets using this package to build models and refine these models to get the better results.


Please sign in to download the files of this thesis.