Bias in random forest variable importance measures: Illustrations, sources and a solution

Strobl, Carolin; Boulesteix, Anne‐Laure; Zeileis, Achim; Hothorn, Torsten

doi:10.1186/1471-2105-8-25

Cited by 2,849 publications

(2,197 citation statements)

References 29 publications

Supporting

Mentioning

2,167

Contrasting

Unclassified

Order By: Relevance

“…Classification and regression trees examine the degree to which factors predict a dependent variable, and determine the relative importance of individual factors (Olden et al 2008;Strobl et al 2009). Specifically, conditional inference trees utilize an iterative, binary recursive data-partitioning algorithm to examine each variable, searching for the best predictor, splitting the data for the dependent variable into two distinct groups, and then repeating the variable selection until no more significant predictors are found (Hothorn et al 2006).…”

Section: Resultsmentioning

confidence: 99%

Local and landscape drivers of arthropod abundance, richness, and trophic composition in urban habitats

et al. 2013

View full text Add to dashboard Cite

Urban green spaces, such as forest fragments, vacant lots, and community gardens, are increasingly highlighted as biodiversity refuges and are of growing interest to conservation. At the same time, the burgeoning urban garden movement partially seeks to ameliorate problems of food security. Arthropods link these two issues (conservation and food security) given their abundance, diversity, and role as providers of ecosystem services like pollination and pest control. Many previous studies of urban arthropods focused on a Author's personal copy single taxon (e.g. order or family), and examined either local habitat drivers or effects of landscape characteristics. In contrast, we examined both local and landscape drivers of community patterns, and examined differences in abundance, richness, and trophic structure of arthropod communities in urban forest fragments, vacant lots, and community gardens. We sampled ground-foraging arthropods, collected data on 24 local habitat features (e.g., vegetation, ground cover, concrete), and examined land-cover types within 2 km of 12 study sites in Toledo, Ohio. We found that abundance and richness of urban arthropods differed by habitat type and that richness of ants and spiders, in particular, varied among lots, gardens, and forests. Several local and landscape factors correlated with changes in abundance, richness, and trophic composition of arthropods, and different factors were important for specific arthropod groups. Overwhelmingly, local factors were the predominant (80 % of interactions) driver of arthropods in this urban environment. These results indicate that park managers and gardeners alike may be able to manage forests and gardens to promote biodiversity of desired organisms and potentially improve ecosystem services within the urban landscape.

show abstract

Section: Resultsmentioning

confidence: 99%

Local and landscape drivers of arthropod abundance, richness, and trophic composition in urban habitats

et al. 2013

View full text Add to dashboard Cite

show abstract

“…Discussion of the algorithm, associated metrics and uses of RF in ecology is provided by Cutler et al [31]. We used the 'randomForests' package implemented in R [34] to run models and the 'cforest' function in package 'party' to obtain unbiased variable importance estimates to corroborate variable selection [35].…”

Section: Methodsmentioning

confidence: 99%

Integrating species traits with extrinsic threats: closing the gap between predicting and preventing species declines

et al. 2010

View full text Add to dashboard Cite

In studies of extinction risk, it is often insufficient to conclude that species with narrow ranges or small clutch sizes require prioritized protection. To improve conservation outcomes, we also need to know which threats interact with these traits to endanger some species but not others. In this study, we integrated the spatial patterns of key threats to Australian amphibians with species' ecological/life-history traits to both predict declining species and identify their likely threats. In addition to confirming the importance of previously identified traits (e.g. narrow range size), we find that extrinsic threats (primarily the disease chytridiomycosis and invasive mosquitofish) are equally important and interact with intrinsic traits (primarily ecological group) to create guild-specific pathways to decline in our model system. Integrating the spatial patterns of extrinsic threats in extinction risk analyses will improve our ability to detect and manage endangered species in the future, particularly where data deficiency is a problem.

show abstract

“…The method injects randomness to guarantee that trees in the forest are different. This somewhat counterintuitive strategy turns out to perform very well compared to many other classifiers, including Strobl et al 2007). Before we classified the Bochanski2007b M dwarf template using RF, we divided each spectrum from 6000 Å to 9000 Å into 600 regions, with each region covering 5 Å.…”

Section: Spectral Typesmentioning

confidence: 99%

M Dwarf Catalog of the Lamost Pilot Survey

Luo

Song

et al. 2014

View full text Add to dashboard Cite

We present a spectroscopic catalog of 58,360 M dwarfs from the Large Sky Area Multi-Object Fiber Spectroscopic Telescope pilot survey. For each spectrum in the catalog, spectral subtype, radial velocity, Hα equivalent width, a number of prominent molecular band indices, and the metal-sensitive parameter ζ are provided. We use the Sloan Digital Sky Survey Data Release 7 Spectroscopic M dwarf catalog to verify the precision of our methods of classifying the spectral types and measuring the radial velocities. The magnetic activity properties of M dwarfs are also traced by Hα emission lines. The molecular band indices included in this catalog are sensitive to temperature or metallicity, and can be used for further study of the physical properties of M dwarfs. This M dwarf catalog is available on the Web site http://sciwiki.lamost.org/MCatalogPilot/.

show abstract

Bias in random forest variable importance measures: Illustrations, sources and a solution

Cited by 2,849 publications

References 29 publications

Local and landscape drivers of arthropod abundance, richness, and trophic composition in urban habitats

Local and landscape drivers of arthropod abundance, richness, and trophic composition in urban habitats

Integrating species traits with extrinsic threats: closing the gap between predicting and preventing species declines

M Dwarf Catalog of the Lamost Pilot Survey

Contact Info

Product

Resources

About