Visual feature engineering

Abstract

Feature engineering is a key concept in machine learning describing the process of defining the characteristics of an observed phenomenon in a way that makes it usable by an algorithm (e.g., [3]). This process often includes domain knowledge to make the features, as well as the results of the algorithms, meaningful in the respective application area. In data analysis generally, including visual data analysis, the obtained results or insights are often dependent on the employed analysis method as well as the parameters and their imensions used. A simple but well-known example is the modifiable area unit problem [5]. Depending on the size and form of the spatial units chosen to aggregate the data, different visualizations and potentially interpretations of the information may result. In some cases, the chosen methods or algorithms and their parameters can be argued to be the right ones to support a specific analysis task, in other cases a sensitivity analysis may be helpful in determining the optimal values. Additionally, visual analytics, allowing tight integration of the interaction with the methods and parameters and the visualizations, has the potential to support the evaluation of the right or sensible analysis method and its parameters as well as to provide provenance information for the finally employed approach.