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).
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. If you use the document you create for this class for any purpose please cite them as they have been very generous in allowing their data to be used for this kind of assignment.
The goal of this project is to build prediction model for the manner in which they did the exercise. This is the “classe” variable in the training set. The output of the project is going to be a report describing how the model has been built, how was the cross validation used, the expected out of sample error, and to explain the choices. Finally, the prediction model will be used to predict 20 different test cases (provided).
Loading of necessary packages for the analysis and obtaining of data. Data will be stored under ‘train’ and ‘test’, and missing values are labeled as ‘NA’.
library(caret)
library(randomForest)
library(rpart)
set.seed(123)
train <- read.csv("https://d396qusza40orc.cloudfront.net/predmachlearn/pml-training.csv", na.strings=c("NA", "", "#DIV/0!"), stringsAsFactors = TRUE)
test <- read.csv("https://d396qusza40orc.cloudfront.net/predmachlearn/pml-testing.csv", na.strings=c("NA", "", "#DIV/0!"), stringsAsFactors = TRUE)We can get an overview of the data:
dim(train)## [1] 19622 160dim(test)## [1] 20 160The first, index column, and the timestamp columns are not relevant, so we can exclude them. Furthermore, since the exploratory data showed that many of the columns are almost entirely filled with NAs (sometimes almost 100%), we can list those columns and exclude them also from further analysis.
colMeans((is.na(train)))[colMeans((is.na(train)))>0.8]## kurtosis_roll_belt kurtosis_picth_belt kurtosis_yaw_belt
## 0.9798186 0.9809398 1.0000000
## skewness_roll_belt skewness_roll_belt.1 skewness_yaw_belt
## 0.9797676 0.9809398 1.0000000
## max_roll_belt max_picth_belt max_yaw_belt
## 0.9793089 0.9793089 0.9798186
## min_roll_belt min_pitch_belt min_yaw_belt
## 0.9793089 0.9793089 0.9798186
## amplitude_roll_belt amplitude_pitch_belt amplitude_yaw_belt
## 0.9793089 0.9793089 0.9798186
## var_total_accel_belt avg_roll_belt stddev_roll_belt
## 0.9793089 0.9793089 0.9793089
## var_roll_belt avg_pitch_belt stddev_pitch_belt
## 0.9793089 0.9793089 0.9793089
## var_pitch_belt avg_yaw_belt stddev_yaw_belt
## 0.9793089 0.9793089 0.9793089
## var_yaw_belt var_accel_arm avg_roll_arm
## 0.9793089 0.9793089 0.9793089
## stddev_roll_arm var_roll_arm avg_pitch_arm
## 0.9793089 0.9793089 0.9793089
## stddev_pitch_arm var_pitch_arm avg_yaw_arm
## 0.9793089 0.9793089 0.9793089
## stddev_yaw_arm var_yaw_arm kurtosis_roll_arm
## 0.9793089 0.9793089 0.9832841
## kurtosis_picth_arm kurtosis_yaw_arm skewness_roll_arm
## 0.9833860 0.9798695 0.9832331
## skewness_pitch_arm skewness_yaw_arm max_roll_arm
## 0.9833860 0.9798695 0.9793089
## max_picth_arm max_yaw_arm min_roll_arm
## 0.9793089 0.9793089 0.9793089
## min_pitch_arm min_yaw_arm amplitude_roll_arm
## 0.9793089 0.9793089 0.9793089
## amplitude_pitch_arm amplitude_yaw_arm kurtosis_roll_dumbbell
## 0.9793089 0.9793089 0.9795638
## kurtosis_picth_dumbbell kurtosis_yaw_dumbbell skewness_roll_dumbbell
## 0.9794109 1.0000000 0.9795128
## skewness_pitch_dumbbell skewness_yaw_dumbbell max_roll_dumbbell
## 0.9793599 1.0000000 0.9793089
## max_picth_dumbbell max_yaw_dumbbell min_roll_dumbbell
## 0.9793089 0.9795638 0.9793089
## min_pitch_dumbbell min_yaw_dumbbell amplitude_roll_dumbbell
## 0.9793089 0.9795638 0.9793089
## amplitude_pitch_dumbbell amplitude_yaw_dumbbell var_accel_dumbbell
## 0.9793089 0.9795638 0.9793089
## avg_roll_dumbbell stddev_roll_dumbbell var_roll_dumbbell
## 0.9793089 0.9793089 0.9793089
## avg_pitch_dumbbell stddev_pitch_dumbbell var_pitch_dumbbell
## 0.9793089 0.9793089 0.9793089
## avg_yaw_dumbbell stddev_yaw_dumbbell var_yaw_dumbbell
## 0.9793089 0.9793089 0.9793089
## kurtosis_roll_forearm kurtosis_picth_forearm kurtosis_yaw_forearm
## 0.9835898 0.9836408 1.0000000
## skewness_roll_forearm skewness_pitch_forearm skewness_yaw_forearm
## 0.9835389 0.9836408 1.0000000
## max_roll_forearm max_picth_forearm max_yaw_forearm
## 0.9793089 0.9793089 0.9835898
## min_roll_forearm min_pitch_forearm min_yaw_forearm
## 0.9793089 0.9793089 0.9835898
## amplitude_roll_forearm amplitude_pitch_forearm amplitude_yaw_forearm
## 0.9793089 0.9793089 0.9835898
## var_accel_forearm avg_roll_forearm stddev_roll_forearm
## 0.9793089 0.9793089 0.9793089
## var_roll_forearm avg_pitch_forearm stddev_pitch_forearm
## 0.9793089 0.9793089 0.9793089
## var_pitch_forearm avg_yaw_forearm stddev_yaw_forearm
## 0.9793089 0.9793089 0.9793089
## var_yaw_forearm
## 0.9793089train <- train[, colMeans(is.na(train))<0.8]
train <- train[, -(3:7)]
train <- train[, -1]
train <- train[, colMeans(is.na(train))<0.8]In order to be in line, the test set also undertakes the same filters.
test <- test[,intersect(colnames(train), colnames(test))]After obtaining tidy data, train data is split into train data (trainSet) with 80% of data, and the rest 20% to the test data (testSet).
partit <- createDataPartition(train$classe, p=0.8, list=FALSE)
trainSet <- train[partit, ]
testSet <- train[-partit, ]The first testing model is the Decision tree from rpart package:
fit1 <- rpart(classe ~ ., data=trainSet, method="class")
predictionsDT <- predict(fit1, testSet, type = "class")
confm1 <- confusionMatrix(predictionsDT, testSet$classe)
confm1## Confusion Matrix and Statistics
##
## Reference
## Prediction A B C D E
## A 1014 150 8 68 24
## B 28 437 60 28 49
## C 23 75 555 105 90
## D 34 68 37 402 44
## E 17 29 24 40 514
##
## Overall Statistics
##
## Accuracy : 0.7448
## 95% CI : (0.7309, 0.7584)
## No Information Rate : 0.2845
## P-Value [Acc > NIR] : < 2.2e-16
##
## Kappa : 0.6759
##
## Mcnemar's Test P-Value : < 2.2e-16
##
## Statistics by Class:
##
## Class: A Class: B Class: C Class: D Class: E
## Sensitivity 0.9086 0.5758 0.8114 0.6252 0.7129
## Specificity 0.9109 0.9479 0.9095 0.9442 0.9656
## Pos Pred Value 0.8022 0.7259 0.6545 0.6872 0.8237
## Neg Pred Value 0.9616 0.9030 0.9580 0.9278 0.9373
## Prevalence 0.2845 0.1935 0.1744 0.1639 0.1838
## Detection Rate 0.2585 0.1114 0.1415 0.1025 0.1310
## Detection Prevalence 0.3222 0.1535 0.2162 0.1491 0.1591
## Balanced Accuracy 0.9098 0.7618 0.8605 0.7847 0.8393From the output it is possible to see that the accuracy of the model is 0.72, so there is room for improvement.
The second model is Random Forest from randomForest package:
fit2 <- randomForest(classe ~ ., data=trainSet)
predictionRF <- predict(fit2, testSet, type = "class")
confm2 <- confusionMatrix(predictionRF, testSet$classe)
confm2## Confusion Matrix and Statistics
##
## Reference
## Prediction A B C D E
## A 1116 5 0 0 0
## B 0 754 3 0 0
## C 0 0 681 4 0
## D 0 0 0 639 6
## E 0 0 0 0 715
##
## Overall Statistics
##
## Accuracy : 0.9954
## 95% CI : (0.9928, 0.9973)
## No Information Rate : 0.2845
## P-Value [Acc > NIR] : < 2.2e-16
##
## Kappa : 0.9942
##
## Mcnemar's Test P-Value : NA
##
## Statistics by Class:
##
## Class: A Class: B Class: C Class: D Class: E
## Sensitivity 1.0000 0.9934 0.9956 0.9938 0.9917
## Specificity 0.9982 0.9991 0.9988 0.9982 1.0000
## Pos Pred Value 0.9955 0.9960 0.9942 0.9907 1.0000
## Neg Pred Value 1.0000 0.9984 0.9991 0.9988 0.9981
## Prevalence 0.2845 0.1935 0.1744 0.1639 0.1838
## Detection Rate 0.2845 0.1922 0.1736 0.1629 0.1823
## Detection Prevalence 0.2858 0.1930 0.1746 0.1644 0.1823
## Balanced Accuracy 0.9991 0.9962 0.9972 0.9960 0.9958The accuracy is now 0.9951, which is a significant improvement. The out of sample error would then be 0.0049 (1-accuracy).
Finally, the predicted values are:
predict <- predict(fit2, test, type = "class")
predict## 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