Machine Learning Course Write-Up
Courtney D. Shelley
September 21, 2014
Introduction
Six young health participants were asked to perform one set of 10 repetitions of the Unilateral Dumbbell Biceps Curl in five different fashions: exactly according to the specification (Class A), throwing the elbows to the front (Class B), lifting the dumbbell only halfway (Class C), lowering the dumbbell only halfway (Class D) and throwing the hips to the front (Class E).
Data sensors were placed on participants’ arm, forearm, belt, and dumbbell. Measurements were assessed in the x-, y-, and z- axes.
Read more: http://groupware.les.inf.puc-rio.br/har#literature#ixzz3DzMBhI3I
Data
The data for this project come from this source: http://groupware.les.inf.puc-rio.br/har.
library(caret); library(stats)
#### Training data set
fileURL <- "https://d396qusza40orc.cloudfront.net/predmachlearn/pml-training.csv"
download.file(fileURL, destfile = "trainingData.csv", method="curl")
training <- read.csv("trainingData.csv", header=TRUE)
dim(training)## [1] 19622 160#### Test data set
fileURL2 <- "https://d396qusza40orc.cloudfront.net/predmachlearn/pml-testing.csv"
download.file(fileURL2, destfile = "testingData.csv", method="curl")
testing <- read.csv("testingData.csv", header =TRUE)
dim(testing)## [1] 20 160set.seed(12345)
inTrain <- createDataPartition(y=training$classe, p=0.7, list=FALSE)
model<-training[inTrain,]
x.val<-training[-inTrain,]
dim(model); dim(x.val)## [1] 13737 160## [1] 5885 160The training data set consists of 19,622 observations on 160 variables. The testing set consists of 20 observations on 160 variables. Training data was split into a model-building set and a cross-validation set to test the model prior to deployment on the testing set. 70% of training observations were randomly assigned to the model category (13737 observations) and 30% of training observations were randomly assigned to the cross-validating set (5885 observations).
Model Creation
Isolation of important variables
Due to the high number of available predictor variables, Principle Component Analysis (PCA) was used to isolate the most influential predictors. Data was first reduced to 48 predictor variables relating to gyros, acceleration, roll, pitch, yaw, and magnent and processed columns of variation, standard deviation, total, max, and min were omitted.
#### Identify important predictor variables by name
gyros<-unique(grep("^gyros_", names(model), value=TRUE)) ## Retain only columns beginning in gyros_
accel<-unique(grep("^accel_", names(model), value=TRUE)) ## Retain only columns beginning in accel_
roll<-unique(grep("^roll_", names(model), value=TRUE)) ## Retain only columns beginning in roll_
pitch<-unique(grep("^pitch_", names(model), value=TRUE)) ## Retain only columns beginning in pitch_
yaw<-unique(grep("^yaw_", names(model), value=TRUE)) ## Retain only columns beginning in yaw_
magnet<-unique(grep("^magnet_", names(model), value=TRUE)) ## Retain only columns beginning in magnet_
greps<-c(gyros, accel, roll, pitch, yaw, magnet) ## Concatenate retained column names
data<-model[, greps] ##Subset model to only columns contained in greps and
#### Identify important predictor variables by Principle Component Analysis
######## I got this from http://strata.uga.edu/6370/lecturenotes/principalComponents.html
data.pca <- prcomp(data, retx=TRUE, center=TRUE, scale=TRUE)
sd <- data.pca$sdev
loadings <- data.pca$rotation # indicate positive correlations between variables and PCs
rownames(loadings) <- colnames(data)
scores <- data.pca$x
### Scree Plot of PCs to determine which to retain ###
var <- sd^2
var.percent <- var/sum(var) *100
barplot(var.percent, xlab="PC", ylab="Percent Variance", main="Scree Plot of Principle Components",
names.arg=1:length(var.percent), las=1,
ylim=c(0,max(var.percent)), col="gray")
abline(h=1/ncol(data)*100, col="red") # Retain PCs above this line
var.percent[1:11]## [1] 16.384 15.087 9.513 8.937 7.385 5.940 4.572 3.910 3.411 2.964
## [11] 2.252sum(var.percent[1:11]) # Retained PCs explain 80% of explained variance## [1] 80.36Model Build: Random Forest Classification
library(randomForest)## randomForest 4.6-10
## Type rfNews() to see new features/changes/bug fixes.###### First time through course I tried Classification Tree. Accuracy of 0.505 and I scored 8/20 on testing set.
data<-model[, c(greps, "classe")] #Training subset with answers
cross.validate<-x.val[, c(greps, "classe")] #Cross Validate subset with answer
test<-testing[,c(greps,"problem_id")] #Testing subset with answers
modFit<-randomForest(classe ~., data=data, ntree=500)
modFit # Error rate of 0.51%##
## Call:
## randomForest(formula = classe ~ ., data = data, ntree = 500)
## Type of random forest: classification
## Number of trees: 500
## No. of variables tried at each split: 6
##
## OOB estimate of error rate: 0.52%
## Confusion matrix:
## A B C D E class.error
## A 3904 1 0 0 1 0.000512
## B 11 2643 4 0 0 0.005643
## C 0 16 2376 4 0 0.008347
## D 0 0 23 2227 2 0.011101
## E 0 0 2 7 2516 0.003564varImpPlot(modFit, sort=TRUE, n.var=11, main="Variable Importance")
Model Test by Cross Validation
X.Val<-predict(modFit, newdata=x.val, type="response")
x.valRight<-X.Val==x.val$classe
table(x.valRight, x.val$classe)##
## x.valRight A B C D E
## FALSE 1 14 14 14 4
## TRUE 1673 1125 1012 950 1078Predictions on Test Data
predict<-predict(modFit, newdata=test, type="response")
predictRight<-predict==test$problem_id
table(predict, test$problem_id)##
## predict 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
## A 0 1 0 1 1 0 0 0 1 1 0 0 0 1 0 0 1 0 0 0
## B 1 0 1 0 0 0 0 1 0 0 1 0 1 0 0 0 0 1 1 1
## C 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0
## D 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0
## E 0 0 0 0 0 1 0 0 0 0 0 0 0 0 1 1 0 0 0 0Submission
pml_write_files = function(x) {
n = length(x)
for(i in 1:n){
filename = paste0("problem_id_", i, ".txt")
write.table(x[i], file=filename,quote=FALSE, row.names=FALSE,col.names=FALSE)
}
}
pml_write_files(predict)