Correlation and variable importance in random forests

Gregorutti, Baptiste; Michel, Bertrand; Saint‐Pierre, Philippe

doi:10.1007/s11222-016-9646-1

Cited by 656 publications

(446 citation statements)

References 36 publications

Supporting

Mentioning

359

Contrasting

Unclassified

Order By: Relevance

“…Neither can we discuss the issue of variable importance stability in detail [36,47,48]. We rather present a study focusing on the application of the RF approach in the light of forest protection issues.…”

Section: Discussionmentioning

confidence: 99%

Improved Outbreak Prediction for Common Pine Sawfly (Diprion pini L.) by Analyzing Floating ‘Climatic Windows’ as Keys for Changes in Voltinism

Möller¹,

Hentschel²,

Wenning³

et al. 2017

Forests

View full text Add to dashboard Cite

Abstract:The biology and population dynamics of pine sawfly Diprion pini L. are extremely complex and variable. Among other factors, climatic conditions determine the potential for mass outbreaks of the species. In this paper, we investigate this influence and describe a statistical approach to identify responsible climatic variables in floating time windows, thus identifying the factors responsible for the transition from latency to outbreak events. Analyses were built upon a data base comprising outbreak events and fine-scaled climatic data for the period 2002-2016 for a model region in the state of Brandenburg, Germany. By applying Random Forest statistic classification analyses, we isolated a set of four variables. They cover precipitation, temperature, and potential evapotranspiration in distinct periods during the current and the previous year. These periods are not fixed in their position but attached to the floating phenological date of bud burst of the host species Pinus sylvestris L. The complete set of variables was able to distinguish forests likely to be defoliated from those not threatened at high probabilities (95% true-positive rate, 98% true-negative rate). The identified climatic windows offer insights into population dynamics in the study region, support adjustments in current monitoring algorithms, and indicate starting points for further investigations covering other regions or different years.

show abstract

Section: Discussionmentioning

confidence: 99%

Improved Outbreak Prediction for Common Pine Sawfly (Diprion pini L.) by Analyzing Floating ‘Climatic Windows’ as Keys for Changes in Voltinism

Möller¹,

Hentschel²,

Wenning³

et al. 2017

Forests

View full text Add to dashboard Cite

show abstract

“…This provides additional flexibility to the RF algorithm (Ma et al, 2006) and allows for favoring sensitivity or specificity to different classes. A variable can be considered a strong predictor when permuting it increases the prediction error (Gregorutti et al, 2017), therefore it's importance I V can be defined as:…”

Section: Evaluation Criteriamentioning

confidence: 99%

Spatial Downscaling of Alien Species Presences Using Machine Learning

2017

View full text Add to dashboard Cite

Spatially explicit assessments of alien species environmental and socio-economic impacts, and subsequent management interventions for their mitigation, require large scale, high-resolution data on species presence distribution. However, these data are often unavailable. This paper presents a method that relies on Random Forest (RF) models to distribute alien species presence counts at a finer resolution grid, thus achieving spatial downscaling. A bootstrapping scheme is designed to account for sub-setting uncertainty, and subsets are used to train a sufficiently large number of RF models. RF results are processed to estimate variable importance and model performance. The method is tested with an ∼8 × 8 km 2 grid containing floral alien species presence and several potentially exploratory indices of climatic, habitat, land use, and soil property covariates for the Mediterranean island of Crete, Greece. Alien species presence is aggregated at 16 × 16 km 2 and used as a predictor of presence at the original resolution, thus simulating spatial downscaling. Uncertainty assessment of the spatial downscaling of alien species' occurrences was also performed and true/false presences and absences were quantified. The approach is promising for downscaling alien species datasets of larger spatial scale but coarse resolution, where the underlying environmental information is available at a finer resolution. Furthermore, the RF architecture allows for tuning toward operationally optimal sensitivity and specificity, thus providing a decision support tool for designing a resource efficient alien species census.

show abstract

“…Although the effect of the correlations on these measures has been studied recently (see Archer and Kimes, 2008;Strobl et al, 2009;Nicodemus and Malley, 2009;Nicodemus et al, 2010;Nicodemus, 2011;Auret and Aldrich, 2011;Tolosi and Lengauer, 2011;Grömping, 2009;Gregorutti et al, 2013), there is not yet a consensus on the interpretation of the importance measures when the predictors are correlated and on what the effect of this correlation is on the importance measure.…”

Section: Variable Importancementioning

confidence: 99%

Evaluation of the satellite-based Global Flood Detection System for measuring river discharge: influence of local factors

Revilla-Romero

Thielen

Salamon

et al. 2014

Hydrol. Earth Syst. Sci.

View full text Add to dashboard Cite

Abstract. One of the main challenges for global hydrological modelling is the limited availability of observational data for calibration and model verification. This is particularly the case for real-time applications. This problem could potentially be overcome if discharge measurements based on satellite data were sufficiently accurate to substitute for groundbased measurements. The aim of this study is to test the potentials and constraints of the remote sensing signal of the Global Flood Detection System for converting the flood detection signal into river discharge values.The study uses data for 322 river measurement locations in Africa, Asia, Europe, North America and South America. Satellite discharge measurements were calibrated for these sites and a validation analysis with in situ discharge was performed. The locations with very good performance will be used in a future project where satellite discharge measurements are obtained on a daily basis to fill the gaps where real-time ground observations are not available. These include several international river locations in Africa: the Niger, Volta and Zambezi rivers.Analysis of the potential factors affecting the satellite signal was based on a classification decision tree (random forest) and showed that mean discharge, climatic region, land cover and upstream catchment area are the dominant variables which determine good or poor performance of the measurement sites. In general terms, higher skill scores were obtained for locations with one or more of the following characteristics: a river width higher than 1 km; a large floodplain area and in flooded forest, a potential flooded area greater than 40 %; sparse vegetation, croplands or grasslands and closed to open and open forest; leaf area index > 2; tropical climatic area; and without hydraulic infrastructures. Also, locations where river ice cover is seasonally present obtained higher skill scores. This work provides guidance on the best locations and limitations for estimating discharge values from these daily satellite signals.

show abstract

Correlation and variable importance in random forests

Cited by 656 publications

References 36 publications

Improved Outbreak Prediction for Common Pine Sawfly (Diprion pini L.) by Analyzing Floating ‘Climatic Windows’ as Keys for Changes in Voltinism

Improved Outbreak Prediction for Common Pine Sawfly (Diprion pini L.) by Analyzing Floating ‘Climatic Windows’ as Keys for Changes in Voltinism

Spatial Downscaling of Alien Species Presences Using Machine Learning

Evaluation of the satellite-based Global Flood Detection System for measuring river discharge: influence of local factors

Contact Info

Product

Resources

About