Packages

library(caTools)
library(rpart)
library(rpart.plot)
library(ROCR)
library(caret)
library(e1071)
install.packages("rattle")
install.packages("pROC")
library(rattle)
library(klaR)

library(foreign)
library(nnet)
library(ggplot2)
library(reshape2)

The goal of your project is to predict the manner in which they did the exercise

This is the "classe" variable in the training set. You may use any of the other

variables to predict with.

You should create a report describing how you built your model, how you used cross

validation, what you think the expected out of sample error is, and why you made the

choices you did. You will also use your prediction model to predict 20 different test cases.

yaw, role, pitch

forearm, dumbbell, belt,

min, max, avg, amplitude, kurtosis, skewness

setwd("~/Desktop/K's/Training/Machine Learning John Hopkins/Project")

Read in data

train = read.csv("pml-training.csv",sep=",",stringsAsFactors=FALSE)

train = read.csv("pml-training.csv")

head(train)
str(train)
summary(train)
table(train$user_name)
table(train$classe)
train$user_name = as.factor(train$user_name)
train$classe = as.factor(train$classe)
train$new_window = as.factor(train$new_window)
table(train$user_name)
train$date = substr(train$cvtd_timestamp,1,10)
table(train$date,train$classe)

Examine the time situation

table(train$cvtd_timestamp)
table(train$raw_timestamp_part_2)

train$date = as.Date(as.character(train$raw_timestamp_part_1), "%d%m%Y")

head(train$date)

Missing

keep = c("X","user_name","cvtd_timestamp",
"new_window","num_window","roll_belt","pitch_belt","yaw_belt","total_accel_belt",
"gyros_belt_x","gyros_belt_y","gyros_belt_z","accel_belt_x",
"accel_belt_y","accel_belt_z","magnet_belt_x","magnet_belt_y","magnet_belt_z",
"roll_arm","pitch_arm","yaw_arm","total_accel_arm","gyros_arm_x","gyros_arm_y",
"gyros_arm_z","accel_arm_x","accel_arm_y","accel_arm_z","magnet_arm_x","magnet_arm_y",
"magnet_arm_z","roll_dumbbell","pitch_dumbbell","yaw_dumbbell",
"total_accel_dumbbell","gyros_dumbbell_x","gyros_dumbbell_y","gyros_dumbbell_z","accel_dumbbell_x",
"accel_dumbbell_y","accel_dumbbell_z","magnet_dumbbell_x","magnet_dumbbell_y","magnet_dumbbell_z",
"roll_forearm","pitch_forearm","yaw_forearm","total_accel_forearm",
"gyros_forearm_x","gyros_forearm_y","gyros_forearm_z","accel_forearm_x","accel_forearm_y",
"accel_forearm_z","magnet_forearm_x","magnet_forearm_y","magnet_forearm_z","classe")
train = train[,(names(train) %in% keep)]
str(train)
summary(train)

test = read.csv("pml-testing.csv")
test = test[,(names(test) %in% keep)]
test$date = substr(test$cvtd_timestamp,1,10)
head(test)
str(test)
summary(test)
nrow(test)

table(test$user_name)

Partitioning the data

partition = createDataPartition(train$date,p=0.5,list=FALSE)
part_train = train[partition,]
nrow(train); nrow(part_train)
table(part_train$date)
table(part_train$date,part_train$classe)

Check correlation between predictor variables

drops = c("classe","cvtd_timestamp","user_name","X")
cor_var = train[,!(names(train) %in% drops)]
str(cor_var)
cor_var = data.matrix(cor_var)
cor_var
M = abs(cor(cor_var))
diag(M) = 0
M
which(M > 0.8,arr.ind=T)

Multinomial logistic regression

train$classe2 <- relevel(train$classe, ref = "A")
mlogreg <- multinom(classe2 ~ . -X -cvtd_timestamp, data = train)
ncol(train)
summary(mlogreg)

cv <- cv.glmnet(train,classe,family="multinomial",nfolds=50,standardize=FALSE)

CART model

ctrl = trainControl(method="cv",number=50)
CARTmodel = train(classe ~ . -X -cvtd_timestamp -date,method="rpart",data=train,trControl=ctrl)
CARTmodel
print(CARTmodel$finalModel)
plot(CARTmodel$finalModel,uniform=TRUE,main="Classification Tree")
text(CARTmodel$finalModel,use.n=TRUE,all=TRUE,cex=0.8)

fancyRpartPlot(CARTmodel$finalModel)
test_predict = predict(CARTmodel,newdata=test)
length(test_predict)
table(test_predict)
table(test$classe,predict(CARTmodel,newdata=test))

Random Forest model

summary(train)
nrow(train)

ctrl = trainControl(method="cv",number=50)
model_rf = train(classe ~ . -X -cvtd_timestamp -date,method="rf",data=part_train,prox=TRUE,
trControl=ctrl)
model_rf
test_predict_rf = predict(model_rf,newdata=test)
table(test_predict_rf)

Boosting with trees

model_btree = train(classe ~ . -X -cvtd_timestamp -date,method="gbm",data=part_train,verbose=FALSE)
model_btree
pbtree = predict(model_btree,test)
pbtree
table(pbtree)

LDA

ctrl = trainControl(method="cv",number=50)
model_lda = train(classe ~ . -X -cvtd_timestamp -date,data=part_train,method="lda",trControl=ctrl)
model_lda
summary(model_lda)
varImp(model_lda, scale = FALSE)
plda = predict(model_lda,test)
plda
table(plda)

Naive Bayes

ctrl = trainControl(method="cv",number=50)
model_nb = train(classe ~ . -X -cvtd_timestamp -date,data=part_train,method="nb",trControl=ctrl)
model_nb
pnb = predict(model_nb,test)
table(pnb)
pnb

SVM

ctrl = trainControl(method="cv",number=50)
model_svmr = train(classe ~ . -X -cvtd_timestamp -date,data=part_train,method="svmRadial",
trControl=ctrl)
model_svml = train(classe ~ . -X -cvtd_timestamp -date,data=part_train,method="svmLinear",
trControl=ctrl)
model_svm
svm_lin = predict(model_svml,test)
svm_lin
table(svm_lin)
svm_rad = predict(model_svmr,test)
svm_rad
table(svm_rad)

PCA analysis

dropsa = c("classe","cvtd_timestamp","raw_timestamp_part_2","raw_timestamp_part_1","user_name","X",
"classe","new_window")
pca_data = train[,!(names(train) %in% dropsa)]
pca_test_data = test[,!(names(test) %in% dropsa)]
str(pca_data)
prComp = prcomp(pca_data)
plot(prComp$x[,1],prComp$x[,2],col=train$classe)
prComp$rotation ### Get the specific attributes for the variables ###

model_pca = preProcess(pca_data,method="pca",pcaComp=2)
train_pca = predict(model_pca,pca_data)
modelFit_pca = train(train$classe ~ .,method="glm",data=train_pca)
test_pca = predict(model_pca,pca_test_data)
plot(test_pca)
text(test_pca,row.names(test_pca))

head(test_pca)

Functions

inTrain = createDataPartition(y= ,p= ,list= ) ### Create test/training sets ###
training = dataset[inTrain,]

folds = createFolds(y= ,k= ,list=TRUE,returnTrain= ) ### k folds ###
sapply(folds,length)
folds[[1]]

folds = createResample(y= ,times= ,list=TRUE) ### Resampling ###

Time slicing

train(model,data=train,method="glm")
confusionMatrix() ### Sensitivity etc. ###

trainControl() ### cross validation ### ### increase number of samples with large data ###

featurePlot(x= ,y= ,plot= ) ### Scatter matrix plot ###
qq = qplot(x,y,color=group,data=training) ### Scatter plot ###
qq + geom_smooth(method="lm",formula=y ~ x)

qplot(x,color=group,data=training,geom="density") ### Density plot ###
preProcess(,method=c()) ### Standardisation ###
nearZeroVar(data,saveMetrics=TRUE) ### Check out variablility of variables ###