Exploration and refinement of regression tree models with interactive maps and spatial data transformations

ABSTRACT

The problem we address is prediction of expected values of some attribute of spatial objects based on values of other attributes, including the geographic positions. A common approach to obtaining such predictions is regression modelling. It is highly desirable that predictive models are not only accurate but also understandable to the users, which gives preference to simpler models. We propose a set of visualisation techniques and interactive operations that supports exploration, evaluation, refinement and simplification of regression tree models. In particular, the analyst can investigate how the model components and their properties are related to the spatial distribution of the objects, and can make the model better account for the spatial aspect of the data by generating new space-based attributes and supplying them to the model building tool.

RÉSUMÉ

Le problème que nous abordons est la prédiction des valeurs attendues d’un attribut d’un objet spatial à partir des valeurs d’autres attributs, y compris les positions géographiques. Une approche classique pour l’obtention de ces prédictions est d’utiliser un modèle de régression. Il est plus que souhaitable que les modèles prédictifs soient non seulement précis mais aussi compréhensibles pour les utilisateurs, ce qui privilégie les modèles simples. Nous proposons un ensemble de techniques de visualisation et d’opérations interactives qui prennent en charge l’exploration, l’évaluation, le raffinement et la simplification des modèles d’arbres de régression. En particulier, l’analyste peut chercher comment les composants du modèle et leurs propriétés sont liées à la répartition spatiale des objets, ce qui permet de mieux tenir compte de l’aspect spatial des données en générant des attributs sur de nouveaux espaces et en les utilisant dans l’outil de construction de modèle.

KEYWORDS:

• Predictive modelling
• geovisualisation
• analytical cartography
• visual analytics

Notes on contributors

Gennady Andrienko is a lead scientist at Fraunhofer Institute IAIS in Sankt Augustin, Germany, and professor at City University London, UK. He was poster chair of IEEE VAST 2013-2014 and is Paper Chair of IEEE VAST 2015 and 2016. Gennady Andrienko is associate editor of three journals, Information Visualization, IEEE Transactions on Visualization and Computer Graphics, and International Journal of Cartography, and editorial board member of Cartography and Geographic Information Science.

Natalia Andrienko is a lead scientist at Fraunhofer Institute IAIS in Sankt Augustin, Germany, and professor at City University London, UK. She is a Tutorial Chair of IEEE VIS 2015, Workshop Chair of IEEE VIS 2016, and Paper Chair of EuroVA 2016. Natalia Andrienko received best paper awards at AGILE 2006, IEEE VAST 2011 and 2012, EuroVis 2015, honorable mention award at IEEE VAST 2010, VAST challenge awards 2008 and 2014, and best poster awards at ACM SIGSPATAIL 2011 and AGILE 2007 conference.

Alexandr Ryumkin is professor at Tomsk State University, faculty of informatics. His research interests are mathematical modelling, data processing and geoinformatics.

Valery Ryumkin is associate professor at Tomsk State University, faculty of economics. His research interest is mathematical modelling of economy.

Gennady Kravchenko is associate professor at Tomsk State University, faculty of informatics. His research topic is geoinformatics.

Evegeny Tyabaev is associate professor at Tomsk State University, faculty of informatics. His research topic is geoinformatics.

Dmitry Khloptosov is professor at Tomsk State University, faculty of economics. His research interest is real estate analysis.

Svetlana Trofimova is a senior lecturer at Tomsk State University, faculty of informatics. Her research topic is geoinformatics.

Disclosure statement

No potential conflict of interest was reported by the authors.

Additional information

Funding

This work was partially funded by the European Commission Horizon 2020 program, within projects SoBigData (‘Social Big Data Research Infrastructure’), [grant agreement 654024] and VaVeL (‘Variety, Veracity, VaLue: Handling the Multiplicity of Urban Sensors’), [grant agreement 688380].