Practical Machine Learning Course Project

Johns Hopkins University

John Cotter

Introduction

Background from Assignment

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, your 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).

Test Setup

Load Required Librarys and set a seed for repeatability

library(caret)
library(randomForest)
set.seed(27272)

Import Data Sets

training=read.csv(file="../pml-training.csv",head=TRUE,sep=",",na.strings=c("NA","#DIV/0!",""))
testing=read.csv(file="../pml-testing.csv",head=TRUE,sep=",")

Clean Data by dropping first 7 columns and Removing columns with all NAs and zeros

training<-training[,-seq(1:7)]
testing<-testing[,-seq(1:7)]
hasNA<-as.vector(sapply(training[,1:152],function(x) {length(which(is.na(x)))!=0}))
training<-training[,!hasNA]
testing<-testing[,!hasNA]

Test Setup

Divide training data as training (70%) and testing (30%)

I will be testing how much accuracy is lost using PCA

inTrain<-createDataPartition(training$classe, p = 0.7)[[1]]
Traintraining<-training[inTrain,]
Traintesting<-training[-inTrain,]

Preprocess with PCA for both training and testing

preProc<-preProcess(Traintraining[,-53],method="pca")
trainPCA<-predict(preProc,Traintraining[,-53])
trainPCA$classe=Traintraining$classe
testPCA<-predict(preProc,Traintesting[,-53])
testPCA$classe=Traintesting$classe

Training with Random Forests

Full data

fitFullRF<-randomForest(Traintraining$classe ~.,data = Traintraining,importance = TRUE)
predictFullRF<-predict(fitFullRF,Traintesting)
fullCM<-confusionMatrix(predictFullRF,Traintesting$classe)
fullCM$overall

##       Accuracy          Kappa  AccuracyLower  AccuracyUpper   AccuracyNull 
##         0.9941         0.9925         0.9917         0.9959         0.2845 
## AccuracyPValue  McnemarPValue 
##         0.0000            NaN

Training with Random Forests

PCA data

fitpcaRF<-randomForest(trainPCA$classe ~.,data = trainPCA,importance = TRUE)
predictpcaRF<-predict(fitpcaRF,testPCA)
pcaCM <- confusionMatrix(predictpcaRF,testPCA$classe)
pcaCM$overall

##       Accuracy          Kappa  AccuracyLower  AccuracyUpper   AccuracyNull 
##      9.752e-01      9.686e-01      9.709e-01      9.790e-01      2.845e-01 
## AccuracyPValue  McnemarPValue 
##      0.000e+00      8.199e-06

Training Error

In random forests, there is no need for cross-validation or a separate test set to get an unbiased estimate of the test set error. It is estimated internally during the run. However, the error does decrease with the number of trees. The following plot shows the training error vs # of trees.

plot(fitFullRF,main="Error vs # of trees")

plot of chunk unnamed-chunk-1

Final Outcomes

Differences in Accuracy

fullCM$overall[1]-pcaCM$overall[1]

## Accuracy 
##  0.01886

PCA only loses ~1.8% Accuracy but I want to use the full data anyway

finalRF<-randomForest(training$classe ~.,data = training,importance = TRUE)
Answer<-predict(finalRF,testing)
Answer

##  1  2  3  4  5  6  7  8  9 10 11 12 13 14 15 16 17 18 19 20 
##  B  A  B  A  A  E  D  B  A  A  B  C  B  A  E  E  A  B  B  B 
## Levels: A B C D E