JLE

Environnement, Risques & Santé

MENU

Geographic data mining to select variables for identifying adverse environmental factors: Method applied to thyroid tumors during a leukemia remission Volume 13, issue 6, November-December 2014

Figures


  • Figure 1

  • Figure 2

  • Figure 3

  • Figure 4

  • Figure 8

  • Figure 12

  • Figure 5

  • Figure 6

  • Figure 9

  • Figure 10

  • Figure 7

  • Figure 11

Tables

Author
Docteur en géographie, attaché d’enseignement et de recherche
Université d’Aix-Marseille (AMU)
Laboratoire UMR-7300-ESPACE (CNRS)
98, boulevard Edouard Herriot
BP 3209
06204 Nice CEDEX 3
* Tirés à part

The French National Cancer Plans call for the assessment of adverse environmental effects on health. Increased access to numerous databases makes it possible to model all dimensions of the environment with geographic information systems (GIS). However, these systems are not suitable for analyzing the complex and highly multidimensional geographic data sets currently available. Complexity generally increases due to the many spatiotemporal uncertainties, as well as when qualitative and quantitative indicators must be incorporated. A recent variable selection strategy based on random forests provides precisely the flexibility and statistical power necessary to overcome this drawback. The MyVsurfGeo (MVG) method implements this tool for spatial analysis. It assesses the negative impact of multiple environmental factors, modeled by various spatial indicators that characterize a small number of municipalities. We apply the MVG method to thyroid tumors that developed during remission of childhood leukemia and discuss its robustness and its expected contributions.