Pattern visualisation in heat maps

Details

Supervisors

Nijssen, Siegfried

Faculty

Ecole polytechnique de Louvain

Degree label

Master [120] : ingénieur civil en informatique, à finalité spécialisée

Abstract

Heat map is a data visualisation technique that represents the magnitude of individual values within a dataset as a color. This method allows the visualization of large datasets and patterns that might otherwise remain hidden in raw data. It has applications across diverse domains, including finance, biology, social sciences, and more. Since an appropriate order of rows and columns reveals more information, efficient ordering techniques are essential. The iterated ordering framework appears as a reasonably good solution, but it suffers from two main limitations: it only works with binary data and cannot be applied to large data sets. As real-world measurements, such as sensor readings, financial data, biomedical data, and environmental data, naturally manifest as big numerical datasets, those limitations often act as real issues and will be addressed in this thesis. First the framework is adapted for non-binary data sets and its performances are evaluated. Then different preprocessing algorithms are proposed to reduce the data size while keeping a maximum of visual information. They are implemented and discussed.