Course details


Time Series Analysis and Forecasting

WS 2019 Dr. rer. nat. Nico Potyka
B.Sc modules:
CS-BWP-AI - Artificial Intelligence
CS-BWP-NI - Neuroinformatics
KOGW-WPM-KI - Artificial Intelligence
KOGW-WPM-NI - Neuroinformatics
M.Sc modules:
CC-MWP-AI - Artificial Intelligence
CC-MWP-NI - Neuroinformatics
CS-MWP-AI - Artificial Intelligence
CS-MWP-NI - Neuroinformatics
KOGW-MWPM-KI - Major subject Artificial Intelligence
KOGW-MWPM-NIR - Major subject Neuroinformatics and Robotics
Tue: 14-16

Time series data is sequential data that is usually observed with a fixed interval between the observations (e.g. daily, weekly, monthly, etc.). Examples include predicting sales or workload and forecasting the weather or natural disasters. As opposed to common classification and regression tasks, it is usually not reasonable to assume that observations in a time series are independent. Therefore, other ideas have to be studied. It is often reasonable to assume that the data-generating process is composed of a systematic (consisting of components like level, trend and seasonality) and a non-systematic component (noise). Forecasting methods can be roughly divided into model-based and data-driven approaches. While model-based approaches attempt to construct a model for the data-generating process, data-driven approaches apply general-purpose learning methods like neural networks in order to predict future events. Model-based approaches are often more transparent and require less data. Since data is often sparse, they still outperform data-driven approaches in many areas. Furthermore, they can be used to remove trend and seasonality components from the observations, which often improves the performance of data-driven approaches. On the other hand, data-driven approaches can be a valuable tool when the assumptions of model-based approaches are violated. They are also more flexible in incorporating external information like special events (e.g. holidays or large social events that can affect sales and traffic conditions). Furthermore, model-based and data-driven approaches can be combined in order to improve the overall performance. The seminar will consist of lectures introducing basic analysis and forecasting tools (both model-based and data-driven), programming and analysis tasks and discussions. Prerequisites: basic knowledge of Python (basic programming and SciPy stack) and Statistics (random variables, descriptive statistics, regression).