========================================================
Human Activity Recognition - HAR - has emerged as a key research area in the last years and is gaining increasing attention by the pervasive computing research community, especially for the development of context-aware systems. Further details can be found here.
The data they have provided was collected 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.
library(caret)## Loading required package: lattice
## Loading required package: ggplot2
trainingData <- read.csv("~/Courses/Practical Machine Learning//pml-training.csv",
na.strings = c("NA", ""))
testingData <- read.csv("~/Courses/Practical Machine Learning//pml-testing.csv")The goal of this project is to predict the manner in which they did the exercise. The manner has been divided into following classes:
levels(trainingData$classe)## [1] "A" "B" "C" "D" "E"
### Cleaning and setting the data: Removing columns with missing values:
# trainingData <- trainingData[,!sapply(data,function(x) any(is.na(x)))]
trainingData <- trainingData[, which(colSums(is.na(trainingData)) == 0)]Also, we remove any variables(columns) that doesn't seem relative(non-sensor data) to the prediction model:
trainingData <- trainingData[, -grep("timestamp|X|user_name|new_window|num_window",
names(trainingData))]After removing the columns with missing values, we get following number of variables:
names(trainingData)## [1] "roll_belt" "pitch_belt" "yaw_belt"
## [4] "total_accel_belt" "gyros_belt_x" "gyros_belt_y"
## [7] "gyros_belt_z" "accel_belt_x" "accel_belt_y"
## [10] "accel_belt_z" "magnet_belt_x" "magnet_belt_y"
## [13] "magnet_belt_z" "roll_arm" "pitch_arm"
## [16] "yaw_arm" "total_accel_arm" "gyros_arm_x"
## [19] "gyros_arm_y" "gyros_arm_z" "accel_arm_x"
## [22] "accel_arm_y" "accel_arm_z" "magnet_arm_x"
## [25] "magnet_arm_y" "magnet_arm_z" "roll_dumbbell"
## [28] "pitch_dumbbell" "yaw_dumbbell" "total_accel_dumbbell"
## [31] "gyros_dumbbell_x" "gyros_dumbbell_y" "gyros_dumbbell_z"
## [34] "accel_dumbbell_x" "accel_dumbbell_y" "accel_dumbbell_z"
## [37] "magnet_dumbbell_x" "magnet_dumbbell_y" "magnet_dumbbell_z"
## [40] "roll_forearm" "pitch_forearm" "yaw_forearm"
## [43] "total_accel_forearm" "gyros_forearm_x" "gyros_forearm_y"
## [46] "gyros_forearm_z" "accel_forearm_x" "accel_forearm_y"
## [49] "accel_forearm_z" "magnet_forearm_x" "magnet_forearm_y"
## [52] "magnet_forearm_z" "classe"
The variable we are predicting is "classe". Now, we partition the data into training set(70%) and cross-validation set(30%).
set.seed(1)
inTrain <- createDataPartition(y = trainingData$classe, p = 0.7, list = FALSE)
training <- trainingData[inTrain, ]
cv <- trainingData[-inTrain, ]Feature density plot:
featurePlot(x = training[, -53], y = training$classe, plot = "density", auto.key = list(columns = 5))
### Building the prediction model: Rather than using the default training method(bootstrapping), we are using cross-validation method with 5 folds to train the data as it seems to be more efficient and fast.
tctrl <- trainControl(method = "cv", number = 5)
modFit <- train(classe ~ ., data = training, method = "rf", trControl = tctrl)## Loading required package: randomForest
## randomForest 4.6-7
## Type rfNews() to see new features/changes/bug fixes.
Summary of our fitted model:
modFit## Random Forest
##
## 3927 samples
## 52 predictors
## 5 classes: 'A', 'B', 'C', 'D', 'E'
##
## No pre-processing
## Resampling: Cross-Validated (5 fold)
##
## Summary of sample sizes: 3142, 3142, 3142, 3142, 3140
##
## Resampling results across tuning parameters:
##
## mtry Accuracy Kappa Accuracy SD Kappa SD
## 2 1 1 0.008 0.01
## 30 1 1 0.008 0.01
## 50 1 1 0.01 0.02
##
## Accuracy was used to select the optimal model using the largest value.
## The final value used for the model was mtry = 27.
plot(modFit)
We can see from above that we have achieved accuracy of 100% using 5 fold Cross-Validated resampling with just 52 predictors.
As for the prediction statistics on cross-validation set:
prediction <- predict(modFit, newdata = cv)## Loading required package: randomForest
## randomForest 4.6-7
## Type rfNews() to see new features/changes/bug fixes.
confusionMatrix(prediction, cv$classe)## Confusion Matrix and Statistics
##
## Reference
## Prediction A B C D E
## A 1660 21 0 0 0
## B 10 1097 29 0 8
## C 1 20 991 19 7
## D 2 1 6 942 6
## E 1 0 0 3 1061
##
## Overall Statistics
##
## Accuracy : 0.977
## 95% CI : (0.973, 0.981)
## No Information Rate : 0.284
## P-Value [Acc > NIR] : < 2e-16
##
## Kappa : 0.971
## Mcnemar's Test P-Value : 0.000241
##
## Statistics by Class:
##
## Class: A Class: B Class: C Class: D Class: E
## Sensitivity 0.992 0.963 0.966 0.977 0.981
## Specificity 0.995 0.990 0.990 0.997 0.999
## Pos Pred Value 0.988 0.959 0.955 0.984 0.996
## Neg Pred Value 0.997 0.991 0.993 0.996 0.996
## Prevalence 0.284 0.194 0.174 0.164 0.184
## Detection Rate 0.282 0.186 0.168 0.160 0.180
## Detection Prevalence 0.286 0.194 0.176 0.163 0.181
## Balanced Accuracy 0.993 0.977 0.978 0.987 0.990
With an accuracy of 97.7%, our out-of-sample error is about 2.3 %