Visual analytics for time series analysis

Time series analysis is a challenging task performed by domain experts in various fields, such as epidemiology, medical signal processing, neurophysiology, environmental sciences, and some research fields where biological data is analyzed. Time series for these fields are often unequally spaced and occasionally include missing values, while standard methods for time series analysis often require equally spaced time series without missing values. To enable the application of the standard methods for model selection in the time domain, such as the Box-Jenkins methodology, these time series therefore have to be transformed. However, the statistical software tools that implement the methods and models for time series analysis lack the adequate intuitive and interactive visual interfaces to support the user with the transformation, imputation, the seamless integration of this time series modifications into the workflow of model selection, and the workflow itself. The goal of this thesis is to overcome these problems by investigating and identifying appropriate Visual Analytics methods for the problem domain, use the findings to design a Visual Analytics process and implement this process in a prototype. The evaluation of the results is done by applying the prototype to an example time series following defined use case scenarios. The evaluation shows that Visual Analytics is a way to overcome the problems mentioned above and to support the user with interactive visualizations and short feedback cycles in the process of time series transformation and model selection.