Feature Selection with the<b>Boruta</b>Package

Kursa, Miron B.; Rudnicki, Witold R.

doi:10.18637/jss.v036.i11

Cited by 3,758 publications

(2,474 citation statements)

References 9 publications

Supporting

Mentioning

2,458

Contrasting

Unclassified

Order By: Relevance

“…Predictive gene families that segregated significantly between soil layers or among harvesting treatments were identified using random forest analysis (Breiman, 2001) with 1000 trees followed by the Boruta algorithm for feature selection (Kursa and Rudnicki, 2010). Gene families were characterized using information from the CAZy website obtained on 15 August 2014 and previous publications on cellulases (Berlemont and Martiny, 2012) and hemicellulases (Shallom and Shoham, 2003;Zhao et al, 2013).…”

Section: Resultsmentioning

confidence: 99%

Forest harvesting reduces the soil metagenomic potential for biomass decomposition

et al. 2015

View full text Add to dashboard Cite

Soil is the key resource that must be managed to ensure sustainable forest productivity. Soil microbial communities mediate numerous essential ecosystem functions, and recent studies show that forest harvesting alters soil community composition. From a long-term soil productivity study site in a temperate coniferous forest in British Columbia, 21 forest soil shotgun metagenomes were generated, totaling 187 Gb. A method to analyze unassembled metagenome reads from the complex community was optimized and validated. The subsequent metagenome analysis revealed that, 12 years after forest harvesting, there were 16% and 8% reductions in relative abundances of biomass decomposition genes in the organic and mineral soil layers, respectively. Organic and mineral soil layers differed markedly in genetic potential for biomass degradation, with the organic layer having greater potential and being more strongly affected by harvesting. Gene families were disproportionately affected, and we identified 41 gene families consistently affected by harvesting, including families involved in lignin, cellulose, hemicellulose and pectin degradation. The results strongly suggest that harvesting profoundly altered below-ground cycling of carbon and other nutrients at this site, with potentially important consequences for forest regeneration. Thus, it is important to determine whether these changes foreshadow long-term changes in forest productivity or resilience and whether these changes are broadly characteristic of harvested forests.

show abstract

Section: Resultsmentioning

confidence: 99%

Forest harvesting reduces the soil metagenomic potential for biomass decomposition

et al. 2015

View full text Add to dashboard Cite

show abstract

“…The wrapper function of RF identifies relevant parameters by performing multiple runs of the provided classification factors that test the performance of different subsets of the input parameters [45]. The RF function is a suitable alternative for the analysis of soil parameters at different depths, as it can be employed without extensive parameter tuning and returns an estimate of the feature's importance (Z-score) [46]. This parameter selection method is different from other dimensionality reduction techniques, such as PCA or other methods.…”

Section: Selection Of Soil Parameters Using the Random Forest Classifiermentioning

confidence: 99%

Spatial Distribution and Mobility Assessment of Carcinogenic Heavy Metals in Soil Profiles Using Geostatistics and Random Forest, Boruta Algorithm

Shaheen

Iqbal

2018

Sustainability

View full text Add to dashboard Cite

Abstract:In third world countries, industries mainly cause environmental contamination due to lack of environmental policies or oversight during their implementation. The Sheikhupura industrial zone, which includes industries such as tanneries, leather, chemical, textiles, and colour and dyes, contributes massive amounts of untreated effluents that are released directly into drains and used for the irrigation of crops and vegetables. This practice causes not only soil contamination with an excessive amount of heavy metals, but is also considered a source of toxicity in the food chain, i.e., bioaccumulation in plants and ultimately in human body organs. The objective of this research study was to assess the spatial distribution of the heavy metals chromium (Cr), cadmium (Cd), and lead (Pb), at three depths of soil using geostatistics and the selection of significant contributing variables to soil contamination using the Random Forest (RF) function of the Boruta Algorithm. A total of 60 sampling locations were selected in the study area to collect soil samples (180 samples) at three depths (0-15 cm, 15-30 cm, and 60-90 cm). The soil samples were analysed for their physico-chemical properties, i.e., soil saturation, electrical conductivity (EC), organic matter (OM), pH, phosphorus (P), potassium (K), and Cr, Cd, and Pb using standard laboratory procedures. The data were analysed with comprehensive statistics and geostatistical techniques. The correlation coefficient matrix between the heavy metals and the physico-chemical properties revealed that electrical conductivity (EC) had a significant (p ≤ 0.05) negative correlation with Cr, Cd, and Pb. The RF function of the Boruta Algorithm employed soil depth as a classifier and ranked the significant soil contamination parameters (Cr, Cd, Pb, EC, and P) in relation to depth. The mobility factor indicated the leachate percentage of heavy metals at different vertical depths of soil. The spatial distribution pattern of Cr, Cd, and Pb revealed spatial variability regarding subsoil horizons. Significant contamination was discovered near the Deg drain and the Bed Nallah irrigated area that indicated a high Cr topsoil contamination, and in a homogenous pattern in Cd and Pb (p < 0.05). Consequently, different soil management strategies can be adopted in an industrial irrigated area to reduce the contamination load of heavy metals in soil.

show abstract

“…The optimization of the parameters chosen in the RF classifier was according to the overall accuracy and the efficiency ( Figure A2). In order to assess the importance of input variables in the classification, we applied the Boruta algorithm for feature selection [45] via an iteration process (R package 'Boruta'). In the post classification, we manually filled the gaps due to residuals of cloud cover in mountain peaks by digitizing from the high-resolution aerial photos.…”

Section: Land Cover Mapping Post Classification and Land Change Anamentioning

confidence: 99%

Continued Reforestation and Urban Expansion in the New Century of a Tropical Island in the Caribbean

Wang

Gao

2017

Remote Sensing

View full text Add to dashboard Cite

Accurate and timely monitoring of tropical land cover/use (LCLU) changes is urgent due to the rapid deforestation/reforestation and its impact on global land-atmosphere interaction. However, persistent cloud cover in the tropics imposes the greatest challenge and retards LCLU mapping in mountainous areas such as the tropic island of Puerto Rico, where forest transition changed from deforestation to reforestation due to the economy shift from agriculture to industry and service after the 1940s. To improve the LCLU mapping in the tropics and to evaluate the trend of forest transition of Puerto Rico in the new century, we integrated the optical Landsat images with the L-band SAR to map LC in 2010 by taking advantage of the cloud-penetrating ability of the SAR signals. The results showed that the incorporation of SAR data with the Landsat data significantly, although not substantially, enhanced the accuracy of LCLU mapping of Puerto Rico, and the Kappa statistic reached 90.5% from 88.4% without SAR data. The enhancement of mapping by SAR is important for urban and forest, as well as locations with limited optical data caused by cloud cover. We found both forests and urban lands continued expanding in the new century despite the declining population. However, the forest cover change slowed down in 2000-2010 compared to that in 1991-2000. The deforestation rate reduced by 42.1% in 2000-2010, and the reforestation was mostly located in the east and southeast of the island where Hurricane Georges landed and caused severe vegetation damage in 1998. We also found that reforestation increased, but deforestation decreased along the topography slope. Reforestation was much higher within the protected area compared to that in the surroundings in the wet and moist forest zones.

show abstract

Feature Selection with theBorutaPackage

Cited by 3,758 publications

References 9 publications

Forest harvesting reduces the soil metagenomic potential for biomass decomposition

Forest harvesting reduces the soil metagenomic potential for biomass decomposition

Spatial Distribution and Mobility Assessment of Carcinogenic Heavy Metals in Soil Profiles Using Geostatistics and Random Forest, Boruta Algorithm

Continued Reforestation and Urban Expansion in the New Century of a Tropical Island in the Caribbean

Contact Info

Product

Resources

About