Support Vector Machines (SVM)
1 like2,547 views
The document details a training session focused on digital soil organic carbon mapping using support vector machines (SVM) and related methodologies. It discusses the advantages of support vector regression (SVR), parameter tuning, and data import in R for effective modeling and prediction. The training emphasizes techniques for selecting covariates, optimizing model performance, and predicting soil organic carbon effectively.
![Support Vector Machines
Using R for Digital Soil Mapping - McBratney et al, 2016
dat <- read.csv("MKD_RegMatrix1.csv")
names(dat)
[1] "X.1" "id" "Y" "X" "SOC" "BLD"
"CRFVOL" "OCSKGM" "meaERROR" "B04CHE3"
[11] "B07CHE3" "B13CHE3" "B14CHE3" "DEMENV5" "LCEE10" "PRSCHE3"
"SLPMRG5" "TMDMOD3" "TMNMOD3" "TWIMRG5"
[21] "VBFMRG5" "VDPMRG5" "soilmap"
Import data > MKD_RegMatrix.csv](https://image.slidesharecdn.com/ird53-svm-180321133152/85/Support-Vector-Machines-SVM-7-320.jpg)
![Support Vector Machines
names(dat)
COR <- cor(as.matrix(dat[,7]), as.matrix(dat[,-c(1:8)]))
COR
x <- subset(melt(COR), value != 1 | value != NA)
x <- x[with(x, order(-abs(x$value))),]
x[1:25,]
idx <- as.character(x$X2[1:25])
Correlation analysis to select covariates](https://image.slidesharecdn.com/ird53-svm-180321133152/85/Support-Vector-Machines-SVM-8-320.jpg)
![Support Vector Machines
dat2 <- dat[c('OCSKGM', idx),]
names(dat2)
files <- list.files(path = "covs", pattern = "tif$",
+ full.names = TRUE)
COV <- stack(files)
COV <- COV[[idx]]
plot(COV)](https://image.slidesharecdn.com/ird53-svm-180321133152/85/Support-Vector-Machines-SVM-9-320.jpg)
![Support Vector Machines
tuneResult <- tune(svm, OCSKGM ~., data = dat[,c("OCSKGM",
names(COV))],
ranges = list(epsilon = seq(0,1,0.1),
cost = c(.5,1,1.5,2,5,10))
)
# Choose the model with the best combination of epsilon and cost
tunedModel <- tuneResult$best.model
Testing the different values for epsilon and cost!
C is a regularization parameter that controls the trade off
between the achieving a low training error and a low testing
error that is the ability to generalize your classifier to unseen
data.](https://image.slidesharecdn.com/ird53-svm-180321133152/85/Support-Vector-Machines-SVM-10-320.jpg)
Support Vector Machines (SVM)
- 1. FAO- Global Soil Partnership Training on Digital Soil Organic Carbon Mapping 20-24 January 2018 Tehran/Iran Yusuf YIGINI, PhD - FAO, Land and Water Division (CBL) Guillermo Federico Olmedo, PhD - FAO, Land and Water Division (CBL)
- 2. SVM
- 3. Support Vector Machines SVM apply a simple linear method to the data but in a high-dimensional feature space non-linearly related to the input space. It creates a hyperplane through n-dimensional spectral-space. Then, SVM separates numerical data based on a kernel function and parameters (e.g. gamma and cost) that maximize the margin from the closest point to the hyperplane that divides data with the largest possible margin, being the support vectors the points which fall within. Then, linear models are fitted to the support vectors.
- 4. Support Vector Machines A major benefit of using SVR is that it is a non- parametric technique. Unlike SLR, whose results depend on Gauss-Markov assumptions, the output model from SVR does not depend on distributions of the underlying dependent and independent variables. Instead the SVR technique depends on kernel functions. https://www.kdnuggets.com/2017/03/building-regression-models-support-vector-regression.html
- 5. Support Vector Machines Another advantage of SVR is that it permits for construction of a non-linear model without changing the explanatory variables, helping in better interpretation of the resultant model. The basic idea behind SVR is not to care about the prediction as long as the error (ϵ) is less than certain value. https://www.kdnuggets.com/2017/03/building-regression-models-support-vector-regression.html
- 7. Support Vector Machines Using R for Digital Soil Mapping - McBratney et al, 2016 dat <- read.csv("MKD_RegMatrix1.csv") names(dat) [1] "X.1" "id" "Y" "X" "SOC" "BLD" "CRFVOL" "OCSKGM" "meaERROR" "B04CHE3" [11] "B07CHE3" "B13CHE3" "B14CHE3" "DEMENV5" "LCEE10" "PRSCHE3" "SLPMRG5" "TMDMOD3" "TMNMOD3" "TWIMRG5" [21] "VBFMRG5" "VDPMRG5" "soilmap" Import data > MKD_RegMatrix.csv
- 8. Support Vector Machines names(dat) COR <- cor(as.matrix(dat[,7]), as.matrix(dat[,-c(1:8)])) COR x <- subset(melt(COR), value != 1 | value != NA) x <- x[with(x, order(-abs(x$value))),] x[1:25,] idx <- as.character(x$X2[1:25]) Correlation analysis to select covariates
- 9. Support Vector Machines dat2 <- dat[c('OCSKGM', idx),] names(dat2) files <- list.files(path = "covs", pattern = "tif$", + full.names = TRUE) COV <- stack(files) COV <- COV[[idx]] plot(COV)
- 10. Support Vector Machines tuneResult <- tune(svm, OCSKGM ~., data = dat[,c("OCSKGM", names(COV))], ranges = list(epsilon = seq(0,1,0.1), cost = c(.5,1,1.5,2,5,10)) ) # Choose the model with the best combination of epsilon and cost tunedModel <- tuneResult$best.model Testing the different values for epsilon and cost! C is a regularization parameter that controls the trade off between the achieving a low training error and a low testing error that is the ability to generalize your classifier to unseen data.
- 11. Support Vector Machines tuneresult Parameter tuning of ‘svm’: - sampling method: 10-fold cross validation - best parameters: epsilon cost 0.4 5 - best performance: 0.7649055 Plot(tuneResult) The best model is the one with lowest MSE. The darker the region the lower the MSE, which means better the model. In our sample data MSE is lowest at epsilon - 0 and cost – 7.
- 12. Support Vector Machines Plot(tuneResult) Testing the different values for epsilon and cost! C is a regularization parameter that controls the trade off between the achieving a low training error and a low testing error that is the ability to generalize your classifier to unseen data.
- 13. SVM # Use the model to predict the SOC in the covariates space beginCluster() start <- Sys.time() pred <- clusterR(COV, predict, args=list(tunedModel)) print(Sys.time() - start) endCluster() plot(pred)
- 14. SVM # Use the model to predict the SOC in the covariates space beginCluster() start <- Sys.time() pred <- clusterR(COV, predict, args=list(tunedModel)) print(Sys.time() - start) endCluster() plot(pred)