load("workspace2016-01-26")
# The Weightlifting data set as 19622 records with 160 columns.
library(AppliedPredictiveModeling)
library(caret)## Loading required package: lattice## Loading required package: ggplot2dat = read.csv("pml-training.csv")
test = read.csv("pml-testing.csv")
# Several columns in the test set has only NA or mostly NA so we removed those.
test2 = test[ , ! apply( test , 2 , function(x) any(is.na(x)) ) ]
# And we realized it was only necessary to train on those columns in the test set. There were more columns in the training set that would never get used anyway.
keepvar = names(test2)
keepvar = keepvar[-60]
keepvar = c(keepvar,"classe")
# It was important to also remove the first 8 columns from the training and test set because there were counters and timestamps that we did not want to affect the outcome. Leaving those in confused the classifiers which thought they were doing better than they were.
keepvar = keepvar[8:60]
dat2 = dat[, keepvar]
# We trained on 75% and tested on the other 25%. It did not appear to matter where the seed started and still got very close solutions.
inTrain = createDataPartition(y=dat2$classe, p = .75, list=FALSE)
training = dat2[ inTrain,]
testing = dat2[-inTrain,]
# The linear discriminant analysis did not work as well only producting 70% accurcay
# modFit1 = train(classe~ .,data=training, method="lda")
pred = predict(modFit1,testing)## Loading required package: MASS#confusionMatrix(pred, testing$classe)
predict(modFit1,test)## [1] B A B C C E D D A A D A B A E A A B B B
## Levels: A B C D E# We tried it with and without cross-validation and it only increased accuracy slightly. Also performing the centering and standardizing between 0 and 1 did not make a difference in the prediction so we didn't continue to do that.
tc = trainControl("repeatedcv",
number=10,
repeats=10,
classProbs=TRUE,
savePred=T)
# modFit2 = train(classe~ .,data=training, method="lda",trControl=tc,
# preProc=c("center", "scale"))
pred = predict(modFit2,testing)
#table(pred, testing$classe)
#confusionMatrix(pred, testing$classe)
predict(modFit2,test)## [1] B A B C C C D D A A D A B A E A A B B B
## Levels: A B C D E# The boosted trees method worked with high accuracy around 97%
# modFit3 = train(classe~ .,data=training, method="gbm")
# It was helpful to see the importance of each variable. At first we tried very small subsets of important variables. Then we realized we could test all of them. It is interesting that it has different lists of importance each time so we couldnt just use this function to tell us which variables to use.
gbmImp = varImp(modFit3, scale = FALSE)## Loading required package: gbm## Loading required package: survival##
## Attaching package: 'survival'## The following object is masked from 'package:caret':
##
## cluster## Loading required package: splines## Loading required package: parallel## Loaded gbm 2.1.1## Loading required package: plyrpred = predict(modFit3,testing)
confusionMatrix(pred, testing$classe)## Confusion Matrix and Statistics
##
## Reference
## Prediction A B C D E
## A 1378 25 0 0 3
## B 14 908 26 0 6
## C 1 15 812 32 6
## D 1 1 13 769 18
## E 1 0 4 3 868
##
## Overall Statistics
##
## Accuracy : 0.9655
## 95% CI : (0.96, 0.9705)
## No Information Rate : 0.2845
## P-Value [Acc > NIR] : < 2.2e-16
##
## Kappa : 0.9564
## Mcnemar's Test P-Value : 0.0001159
##
## Statistics by Class:
##
## Class: A Class: B Class: C Class: D Class: E
## Sensitivity 0.9878 0.9568 0.9497 0.9565 0.9634
## Specificity 0.9920 0.9884 0.9867 0.9920 0.9980
## Pos Pred Value 0.9801 0.9518 0.9376 0.9589 0.9909
## Neg Pred Value 0.9951 0.9896 0.9894 0.9915 0.9918
## Prevalence 0.2845 0.1935 0.1743 0.1639 0.1837
## Detection Rate 0.2810 0.1852 0.1656 0.1568 0.1770
## Detection Prevalence 0.2867 0.1945 0.1766 0.1635 0.1786
## Balanced Accuracy 0.9899 0.9726 0.9682 0.9742 0.9807predict(modFit3,test)## [1] B A B A A E D B A A B C B A E E A B B B
## Levels: A B C D E# The rpart method did only produced around 65% accuracy for some reason.
# modFit4 = train(classe~ .,data=training, method="rpart",)
pred = predict(modFit4,testing)## Loading required package: rpart#confusionMatrix(pred, testing$classe)
predict(modFit4,test)## [1] C D D C C C D D A A C D C A D D C D C D
## Levels: A B C D E# The random forest method worked the best with possibly more than 97% accuracy.
# modFit5 = train(classe~ .,data=training, method="rf")
pred = predict(modFit5,testing)## Loading required package: randomForest## randomForest 4.6-12## Type rfNews() to see new features/changes/bug fixes.##
## Attaching package: 'randomForest'## The following object is masked from 'package:ggplot2':
##
## marginconfusionMatrix(pred, testing$classe)## Confusion Matrix and Statistics
##
## Reference
## Prediction A B C D E
## A 1395 0 0 0 0
## B 0 949 0 0 0
## C 0 0 855 3 0
## D 0 0 0 801 2
## E 0 0 0 0 899
##
## Overall Statistics
##
## Accuracy : 0.999
## 95% CI : (0.9976, 0.9997)
## No Information Rate : 0.2845
## P-Value [Acc > NIR] : < 2.2e-16
##
## Kappa : 0.9987
## Mcnemar's Test P-Value : NA
##
## Statistics by Class:
##
## Class: A Class: B Class: C Class: D Class: E
## Sensitivity 1.0000 1.0000 1.0000 0.9963 0.9978
## Specificity 1.0000 1.0000 0.9993 0.9995 1.0000
## Pos Pred Value 1.0000 1.0000 0.9965 0.9975 1.0000
## Neg Pred Value 1.0000 1.0000 1.0000 0.9993 0.9995
## Prevalence 0.2845 0.1935 0.1743 0.1639 0.1837
## Detection Rate 0.2845 0.1935 0.1743 0.1633 0.1833
## Detection Prevalence 0.2845 0.1935 0.1750 0.1637 0.1833
## Balanced Accuracy 1.0000 1.0000 0.9996 0.9979 0.9989predict(modFit5,test)## [1] 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# The next step would be to combine all the models and either average the answers or vote with them. But in this case random forests scored 100% on the 20 question validation set so it was not necessary.gbmImp ## gbm variable importance
##
## only 20 most important variables shown (out of 52)
##
## Overall
## roll_belt 2907.6
## pitch_forearm 1529.4
## yaw_belt 1256.5
## magnet_dumbbell_z 940.0
## magnet_dumbbell_y 802.7
## roll_forearm 733.5
## magnet_belt_z 567.2
## gyros_belt_z 453.4
## roll_dumbbell 387.4
## pitch_belt 370.7
## accel_forearm_x 367.4
## gyros_dumbbell_y 321.6
## accel_dumbbell_y 286.1
## magnet_belt_y 228.5
## magnet_forearm_z 211.2
## accel_dumbbell_x 206.7
## yaw_arm 188.4
## accel_belt_z 167.6
## magnet_dumbbell_x 148.7
## magnet_arm_z 144.3Velloso, E.; Bulling, A.; Gellersen, H.; Ugulino, W.; Fuks, H. Qualitative Activity Recognition of Weight Lifting Exercises. Proceedings of 4th International Conference in Cooperation with SIGCHI (Augmented Human ’13) . Stuttgart, Germany: ACM SIGCHI, 2013.
Read more: http://groupware.les.inf.puc-rio.br/har#weight_lifting_exercises#ixzz3yZx0m9BL