Practicle Machine learning project – Predict the wearable activities¶

Introduction¶

Using devices such as Jawbone Up, Nike FuelBand, and Fitbit it is now possible to collect a large amount of data about personal activity relatively inexpensively. These type of devices are part of the quantified self movement – a group of enthusiasts who take measurements about themselves regularly to improve their health, to find patterns in their behavior, or because they are tech geeks. One thing that people regularly do is quantify how much of a particular activity they do, but they rarely quantify how well they do it.

In this project, the goal will be to use data from accelerometers on the belt, forearm, arm, and dumbell of 6 participants. They were asked to perform barbell lifts correctly and incorrectly in 5 different ways. More information is available from the website here: http://groupware.les.inf.puc-rio.br/har (see the section on the Weight Lifting Exercise Dataset).

Data¶

The training data for this project are available here:

https://d396qusza40orc.cloudfront.net/predmachlearn/pml-training.csv

The test data are available here:

https://d396qusza40orc.cloudfront.net/predmachlearn/pml-testing.csv

The data for this project come from this source: http://groupware.les.inf.puc-rio.br/har.

The data can be downloaded using the below R script.

In [ ]:

%%R
downloadFiles<-function(dataURL="", destF="t.csv"){
        if(!file.exists(destF)){
            download.file(dataURL, destF, method="wget")
        }else{
            message("data already downloaded.")
        }
    }

loading training and testing dataset

In [ ]:

%%R
trainURL<-"https://d396qusza40orc.cloudfront.net/predmachlearn/pml-training.csv"
testURL <-"https://d396qusza40orc.cloudfront.net/predmachlearn/pml-testing.csv"

Download traninig files

In [ ]:

%%R
downloadFiles(trainURL, "pml-training.csv")

Download Test File

In [ ]:

%%R
downloadFiles(testURL, "pml-test.csv")

read train and test files into R

In [ ]:

%%R
training <- read.csv("pml-training.csv",na.strings=c("NA",""))
testing <-read.csv("pml-test.csv",na.strings=c("NA",""))

First look of training data

In [ ]:

%%R
dim(training)

In [ ]:

%%R
str(training)

outocme is "classe" variable

In [ ]:

%%R
table(training$classe)

Removing missing values

Get only those variables, for which there is no missing value

In [ ]:

%%R
var <- names(training)[apply(training,2,function(x) table(is.na(x))[1]==19622)]

In [ ]:

%%R
length(names(training))

In [ ]:

%%R
length(var)

Build the training dataset from these predictor variables.

In [ ]:

%%R
train2<- training[,var]

test dataset no classe variable

In [ ]:

%%R
test2 <- testing[,c(var[-length(var)],names(testing)[length(testing)])]

Discard unuseful predictors. Consider only numeric variables from HAR sensor

In [ ]:

%%R
removeIndex <- grep("timestamp|X|user_name|new_window|num_window",names(train2))
train3 <- train2[,-c(removeIndex, length(train2))]
test3  <- test2[,-c(removeIndex, length(test2))]

Check the near Zero covariates and correlation matrix removing zero covariates

In [ ]:

%%R
nzv <- nearZeroVar(train3, saveMetrics=TRUE)

In [ ]:

%%R
length(names(train3))

In [ ]:

%%R
dim(nzv)

In [ ]:

%%R
head(nzv)

Find the variables which have very less covariance (less significant)

In [ ]:

%%R
nzv[nzv$nzv,]

no rows implies highly correlated covariates.

Generate Correlation plot

In [ ]:

%%R
#install.packages('corrplot',dep=T)
library(corrplot)
corrM <- cor(train3)
corrplot(corrM, method="circle",tl.cex=0.5)

Remove highly correlated variables

highly correlated variables

In [ ]:

%%R
highCorr <- findCorrelation(corrM, cutoff = .75); length(highCorr)

Build a training set by removing highly correlated variables.

In [ ]:

%%R
train4 <- cbind(classe=train2$classe,train3[,-highCorr])

Select the same columns in test dataset

In [ ]:

%%R
test4 <- test3[, -highCorr]        # dataframe of test predictors

Split training dataset into training/testing for model evaulation

In [ ]:

%%R
set.seed(1234)
inTrain = createDataPartition(train4$classe, p = 3/4)[[1]]
trainPart = train4[ inTrain,]
testPart =  train4[-inTrain,]

Use Random Forest algorithm for prediction

In [ ]:

%%R
library(randomForest);

In [ ]:

%%R
rfModel <- randomForest(classe ~ .,data = trainPart,importance = TRUE,ntrees = 500)
print(rfModel)

Plot the model

In [ ]:

%%R
par(mar=c(3,4,4,4))                               
plot(rfModel)

Output Parameters available from the fitted model

In [ ]:

%%R
names(rfModel)

Confusion Matrix

In [ ]:

%%R
rfModel$confusion

In [ ]:

%%R
varImpPlot(rfModel,cex=.5)

Test sample and cross validation

Predict the output

In [ ]:

%%R
out.test <- predict(rfModel,testPart)

Compare the results

In [ ]:

%%R
table(testPart$classe, out.test)

In [ ]:

%%R
out.test<-predict(rfModel,test4) 
out.test[1:20]

k fold cross validataion, which takes too much computing time

Preprocessing the data.

PCA with threshold 0.80, two misclassified in the first 20.

In [ ]:

%%R
preProc <- preProcess(trainPart[,-1], method="pca", thresh=0.8)
trainPC <- predict(preProc, trainPart[,-1])

Generate randomForest Model

In [ ]:

%%R
modPC   <- randomForest(trainPart$classe~., data=trainPC, importance=TRUE, ntree=10)

Apply same pre processing in test dataset

In [ ]:

%%R
testPC  <- predict(preProc, testPart[,-1])

Predict results

In [ ]:

%%R
out.testPC <-predict(modPC, newdata=testPC)

In [ ]:

%%R
table(out.testPC, testPart$classe)

Save the output

In [ ]:

%%R
answers<- as.vector(out.test[1:20]);answers

In [ ]:

%%R
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(answers)