Practical Machine Learning
Prediction Assignment Writeup
Juan Carlos Bretal Fernandez, December 2014
This is my analisys and solution to the problem. The Random Forest method is the best for this large dataset with too many predictors. Its necesary aplly a preprocesing. The `pml-testing.csv´ dont have all the columns and then i toke the option of remove all the "unused" columns. The train is SLOW but very accurate. See Confusion Matrix a validation at the bottom of this file.
#load data and libraries
library(caret)
library(doMC)
registerDoMC(cores = 2)
set.seed(2)
datos = read.csv("~/Descargas/pml-training.csv")
training <- subset(datos,select=-c(X,
user_name,
raw_timestamp_part_1,
raw_timestamp_part_2,
cvtd_timestamp))
test <- read.csv("~/Descargas/pml-testing.csv")
testing <- subset(test,select=-c(X,
user_name,
raw_timestamp_part_1,
raw_timestamp_part_2,
cvtd_timestamp,
problem_id))
# comparative of unused columns training vs test set
nzvTrain <- nearZeroVar(training)
nzvTest <- nearZeroVar(testing)
nzvTest %in% nzvTrain
filteredTraining <- training[, -nzvTest]# removing unused columns
dim(filteredTraining)
set.seed(2)
# creating a partition over the training set to achieve validation set
inTrain = createDataPartition(filteredTraining$classe, p = 0.6)[[1]]
trainFit = filteredTraining[ inTrain,]
testFit = filteredTraining[-inTrain,]
# fit to random forest with preprocessing
modFit <- train( trainFit$classe ~ . ,method="rf",
allowParallel=TRUE ,
data=trainFit ,
preProcess = c("center", "scale"))
print(modFit$finalModel)
Call:
randomForest(x = x, y = y, mtry = param$mtry, allowParallel = TRUE)
Type of random forest: classification
Number of trees: 500
No. of variables tried at each split: 27
OOB estimate of error rate: 0.38%
Confusion matrix:
A B C D E class.error
A 3347 0 0 0 1 0.0002986858
B 9 2267 3 0 0 0.0052654673
C 0 11 2043 0 0 0.0053554041
D 0 0 13 1915 2 0.0077720207
E 0 0 0 6 2159 0.0027713626
Its a very good acuracy. This will be my prediction for the test set in file pml-testing.csv.
filteredTest <- testing[, -nzvTest]# removing unused columns # adding classe column filteredTest["classe"] <- filteredTraining[dim(filteredTraining)[2]] # prediction over the test set pml-testing.csv prediction <- predict(modFit, newdata=filteredTest) ______________________________________________ prediction [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 ______________________________________________
Prediction over the test set partition of training set for aditional cross validation (40% of training set). Total accuracy of 99,5% which is a great value.
# to calculate accuracy and confusion matrix predict1 <- predict(modFit, testFit) cm <- confusionMatrix(predict1,testFit$classe) Confusion Matrix and Statistics Reference Prediction A B C D E A 2232 8 0 0 0 B 0 1508 11 0 0 C 0 2 1357 9 0 D 0 0 0 1277 7 E 0 0 0 0 1435 Overall Statistics Accuracy : 0.9953 95% CI : (0.9935, 0.9967) No Information Rate : 0.2845 P-Value [Acc > NIR] : < 2.2e-16 Kappa : 0.994 Mcnemar's Test P-Value : NA Statistics by Class: Class: A Class: B Class: C Class: D Class: E Sensitivity 1.0000 0.9934 0.9920 0.9930 0.9951 Specificity 0.9986 0.9983 0.9983 0.9989 1.0000 Pos Pred Value 0.9964 0.9928 0.9920 0.9945 1.0000 Neg Pred Value 1.0000 0.9984 0.9983 0.9986 0.9989 Prevalence 0.2845 0.1935 0.1744 0.1639 0.1838 Detection Rate 0.2845 0.1922 0.1730 0.1628 0.1829 Detection Prevalence 0.2855 0.1936 0.1744 0.1637 0.1829 Balanced Accuracy 0.9993 0.9958 0.9951 0.9960 0.9976
Conclusion: the Random Forest with preprocesing its the more accurate method, but with the 60% of the training set tooks too many to do it daily, may be other way to do it faster. Perhaps a drastical reduction of predictors to the most importants only. And reducing the dataset size to a value less than 11Mb for the training set.
