Multidimensional Data Visualization

As the scale and complexity of data sets increase, not only the number of data points becomes larger, but also the number of features (or dimensions) that describe them. Since human beings are only naturally able to visualize up to 3 dimensions at a time, visualization researchers have designed different abstractions and smart workarounds to be able to analyze patterns and extract insights from multidimensional data. Some examples are Parallel Coordinate Plots, Scatterplot Matrices, or Star Coordinates. These have been successfully applied in a diverse range of application domains, such as Biology, Astronomy, or Social Sciences.

In our group, we focus most of our multidimensional visualization research in one group of techniques: Dimensionality Reduction (DR). In summary, DR methods project the data into 2 or 3 dimensions while retaining some aspects of the original structure of the data, such as global or local distances between points, for example. The results are then usually visualized with enhanced and interactive scatterplots. Every DR method is different than the other, optimizing different loss functions and favoring more local or global aspects of the data. We have applied DR methods to the analysis of multidimensional data from software engineering, education, and topics in academic papers, for example.