Note

Before running the script, set working directory to the one containing Rmd file.

Loading and preprocessing the data

Unzip and load the data into activity dataset.

Discard missing values in defactivity dataset and building the date factor column again with actual values.

unzip(paste(getwd(), "activity.zip", sep = "/"))
activity <- read.csv("activity.csv")
defactivity <- activity[!is.na(activity$steps), ]
defactivity$date <- as.factor(as.Date(defactivity$date))

Histogram of the total number of steps taken each day

Calculate the array of the total number of steps per day (totalstepsperday)

totalstepsperday <- with(defactivity, tapply(steps, date, sum, na.rm = T))

First we plot a barplot of total number of steps per day

par(mfrow = c(1, 1))
barplot(totalstepsperday, space = 0.2, cex.names = 0.5, col = rainbow(61), las = 2,
        cex.axis = 0.8, main = "Total number of steps taken each day")
grid(nx = NA, ny = NULL)

Then we plot the actual histogram of total number of steps per day

hist(totalstepsperday, breaks = seq(from = 0, to = 22500, by = 2500),
     col = rainbow(10), border = "grey", density = 100,
     main = "Histogram of total number of steps per day",
     xlab = "Total number of steps per day")
grid(nx = NA, ny = NULL)

We can check, for example, the number of days with more than 20000 steps (2) regarding the previous plot.

What is mean total number of steps taken per day?

We use the mean and median function on the totalstepsperday array

meanstepsnumber <- as.character(round(mean(totalstepsperday), digits = 2))
medianstepsnumber <- as.character(round(median(totalstepsperday), digits = 2))

The mean number of steps taken per day is 10766.19.

The median number of steps taken per day is 10765.

What is the average daily activity pattern?

Time series plot of the average number of steps taken

dd <- defactivity
dd$interval <- as.factor(dd$interval)
meanstepsperinterval <- with(dd, tapply(steps, interval, mean))
plot(names(meanstepsperinterval), meanstepsperinterval, col = "red", pch = 18,
     xlab = "Interval", ylab = "Mean number of steps",
     main = "Time series plot of the average number of steps taken", type = "l",
     xlim = c(0, 2400))

The 5-minute interval that, on average, contains the maximum number of steps

meanstepsperinterval <- with(defactivity, tapply(steps, interval, mean, na.rm = T))
maxmeanstepsperinterval <- max(meanstepsperinterval)
intervalmaxmeanindex <- which(meanstepsperinterval == maxmeanstepsperinterval)
intervalmaxmean <- names(meanstepsperinterval)[intervalmaxmeanindex]
immSTR <- strsplit(as.character(intervalmaxmean), "")[[1]]
immH <- paste(ifelse(length(immSTR)>2, immSTR[1], "0"), " H", sep = "")
immMd <- ifelse(length(immSTR)>2, immSTR[2], ifelse(length(immSTR)>1, immSTR[1], "0"))
immMu <- ifelse(length(immSTR)>2, immSTR[3], ifelse(length(immSTR)>1, immSTR[2], immSTR[1]))
immM <- paste(immMd, immMu, " mn", sep = "")

intervalmaxmeanHM <- paste(immH, immM, sep = " ")

The 5-minute interval that, on average, contains the maximum number of steps is 835 (8 H 35 mn) - mean total number of steps for this interval : 206.17

Imputing missing values

Number of missing values

nbofmissingdata <- length(activity$steps[which(is.na(activity$steps))])

The number of missing data is 2304.

A quick look on where values are missing. Display a table with the count of missing value per day with missing values.

patternofmissingdata <- with(activity, tapply(steps, date, function(x) {
    length(which(is.na(x)))
}))
tt <- patternofmissingdata[which(patternofmissingdata>0)]
tt
## 2012-10-01 2012-10-08 2012-11-01 2012-11-04 2012-11-09 2012-11-10 
##        288        288        288        288        288        288 
## 2012-11-14 2012-11-30 
##        288        288
nbofdayswithmissingvalues <- length(tt)

There are 8 days with missing values.

Impute missing values

We choose to impute the mean per interval calculated with non empty data to the non available data.

We calculate the mean of total steps per interval

meanstepsperinterval <- with(defactivity, tapply(steps, interval, mean))

Then we create an index of missing values.

imputedactivitymeaninterval <- activity
missingIndex <- which(is.na(imputedactivitymeaninterval["steps"]))

And, using this index, we set each value with appropriate mean value for current interval

for (i in missingIndex) {
    imputedactivitymeaninterval[i, "steps"] <- meanstepsperinterval[as.character(imputedactivitymeaninterval[i, "interval"])]
}

Barplot and histogram of total number of steps per day, with imputed data

totalstepsperdayimp <- with(imputedactivitymeaninterval, tapply(steps, date, sum, na.rm = T))
par(mfrow = c(2, 1))
barplot(totalstepsperdayimp, space = 0.2, cex.names = 0.5, col = rainbow(61), las = 2,
        cex.axis = 0.8, main = "Total number of steps taken each day")
grid(nx = NA, ny = NULL)
hist(totalstepsperdayimp, breaks = seq(from = 0, to = 22500, by = 2500),
     col = rainbow(10), border = "grey", density = 100,
     main = "Histogram of total number of steps per day",
     xlab = "Total number of steps per day")
grid(nx = NA, ny = NULL)

Mean and median with imputed data

meanstepsnumberimp <- as.character(round(mean(totalstepsperdayimp), digits = 2))
medianstepsnumberimp <- as.character(round(median(totalstepsperdayimp), digits = 2))

With imputed missing values, the mean number of steps taken per day is 10766.19. Without considering missing values, the mean number of steps taken per day is 10766.19.

The median number of steps taken per day is 10766.19. Without considering missing values, the median number of steps taken per day is 10765.

With imputed missing value, mean stays the same (obviously), median is slightly higher.

Are there differences in activity patterns between weekdays and weekends?

Add a factor column for weekday / weekend

Instead of using weekdays (depends on language / culture), we use the number of day in week, given by as.POSIXlt(…)$wday

imputedactivitymeaninterval$wkd <- ifelse(as.POSIXlt(imputedactivitymeaninterval$date)$wday %in% 1:5, "weekday", "weekend")
imputedactivitymeaninterval$wkd <- as.factor(imputedactivitymeaninterval$wkd)

Shape the data for lattice plotting, using reshape2 library

library(reshape2)
imputedactivitymeanintervalMelt <- melt(imputedactivitymeaninterval,
                        id = c("wkd", "interval"), measure.vars = c("steps"))
emData <- dcast(imputedactivitymeanintervalMelt, interval + wkd ~ variable, mean)

Plot the two graphs (weekend / weekday) one above the other using lattice library

library(lattice)
g <- xyplot(steps ~ interval | wkd, data = emData, main = "", ylab = "Number of steps",
       xlab = "Interval", cex = 1, col = "blue", horizontal = FALSE, type = "l",
       index.cond = list(c(1, 2)))
g <- update(g, layout = c(1, 2))
plot(g)

There is a difference between weekdays and weekend days : the pike on 8h35 is not as high on weekend days as on weekdays. Another difference is for example the earlier beginning of steps in weekdays (5h).