- Walking Wales: the visualisation challenge
- Word Clouds: Can they be misleading?
- VIPS-Visual Interactive Parameter Steering
- Confirmation bias : do multiple views really help?
- Visual Query Interface for Infrastructure Networks
- MADucator – Matrix Educator
- Concensus Matrix Sort
- Multiscale Multiples: An Overview+Detail Interface for Small Multiples
- Analyzing Image Similarity to Detect Disguised Plagiarism
- Tackling the Multi-Analyst Problem in Soccer: How to Improve Collaborative Pattern Detection in Team Sports
- MDSQ – Quality Assessment of Distance-Preserving Projections
- Internal Quality Measures for Subspace Clusterings
- Exploration of Datasets for Visualizations of High-Dimensional Data
- Search and Visual Exploration of Scientific Literature
- Star Glyph – Optimal Dimension Layout
- Visual Analytics of Molecular Neurochemicals and Biological Markers in Mental Illness
- Ensembles of classifiers: Visual construction of classification models
- Cutting-Down the Complexity of Parallel Coordinates Plots in High-Dimensional Data
- Investigating Analytic Behavior in VA
- Hierarchical Matrix Visualization
- Realisierung und Evaluierung einer stereoskopischen 3D Perspective Wall Umgebung für verlinkte Informationsvisualisierung (Master)
- Combination of Matrices + Graphs
- TreeMap Evaluation
- Visual Analysis of Language Change over Time
- Visual Parameter Space Analysis of Topic Models
|Theoretical (Analytical):||(2 / 5)|
|Practical (Implementation):||(5 / 5)|
|Literature Work:||(3 / 5)|
Machine learning classifiers can be built as a group of several distinct classification models, in such a way that this ensemble of models can produce better classification results than a single model. However, often a crucial step is the selection of an initial set of classification models in order to build an ensemble classifier, and we argue that data visualization can facilitate this process.
The goal of this project is the development of a visual interactive interface to build ensembles of classifiers. This interface should present to the user information about the performance of each single model, so he or she can better decide on how to build the ensemble. At the end, we want to show that our approach provides relevant support to the user in comparison with non-visual approaches, and also in comparison with existing tools that provide any kind of Graphical User Interface (GUI).
- Literature review about classification model ensembles, including scientific papers from the visual analytics and machine learning research communities.
- Analysis of public existing tools w/ GUI for building ensembles of classifiers (e.g. Weka)
- Development of new approaches for the construction of ensembles of classifiers
- Implementation and evaluation of the proposed solution
- Basic knowledge about machine learning classifier algorithms
- Good programming skills in Java
- Motivation to read scientific papers about data visualization and machine learning
Justin Talbot, Bongshin Lee, Ashish Kapoor, and Desney S. Tan. 2009. EnsembleMatrix: Interactive Visualization to Support Machine Learning with Multiple Classiﬁers. In Proceedings of the SIGCHI Conference on Human Factors in Computing Systems (CHI ’09). ACM, New York, NY, USA, 1283–1292.
Ensemble selection in a Nutshell