Exploratory Data Analysis with Fitbit Data

Last updated on May 16, 2021 16 min read Data Viz

I thought it would be awesome to analyze my own data, therefore in June 2019, I started tracking activities with Fitbit. Fitbit isn’t a perfect measure of activities, but it can provide an overall idea of how active you are and how you sleep.

In this post, we are going to take activities and sleep data collected for over a year and perform Exploratory Data Analysis(EDA) to find some fun and interesting insights. We will spend exploration mostly on the steps. Additionally, we will see if activities are correlated. Let’s analyze them together.

First, we start by loading required libraries, and then follow EDA steps below.

library(pacman)
p_load(tidyverse,ggpubr,xts,dygraphs,corrplot,dlookr,ggcorrplot)

1. Get the data

Data was exported directly. Fitbit data can also be accessed via API method.

fitbit_data <- read_csv('fitbit.csv')

2. Cleaning the data

Variables are self explanatory. Some variables are not in correct format, so we need to change them.

#Change variables type
fitbit_data$Date <- as.Date(fitbit_data$Date, format = "%m/%d/%Y")
fitbit_data$Month <- factor(fitbit_data$Month)
fitbit_data$Minutes_REM_Sleep <- as.integer(fitbit_data$Minutes_REM_Sleep)
fitbit_data$Minutes_Light_Sleep <- as.integer(fitbit_data$Minutes_Light_Sleep)
fitbit_data$Minutes_Deep_Sleep <- as.integer(fitbit_data$Minutes_Deep_Sleep)
fitbit_data$Day <- as.factor(fitbit_data$Day)

Since we have enough data points, let’s remove NA values for accurate representation. Additionally, we create a variable “Minutes_Active” using dpylr mutate() function.

fitbit_data1 <- na.omit(fitbit_data) # remove na's
# add a column Minutes_Active
fitbit_data1 <- fitbit_data1 %>% mutate(Minutes_Active = (Minutes_Fairly_Active+Minutes_Lightly_Active+Minutes_Very_Active))
head(fitbit_data1)

## # A tibble: 6 x 18
##   Date       Month Day   Calories_Burned Steps Minutes_Sedenta… Minutes_Lightly…
##   <date>     <fct> <fct>           <dbl> <dbl>            <dbl>            <dbl>
## 1 2019-06-11 Jun   Tue              2867 11699              847              170
## 2 2019-06-13 Jun   Thu              2687  8087              842              205
## 3 2019-06-15 Jun   Sat              3226 13815              699              272
## 4 2019-06-17 Jun   Mon              2895 11348              845              192
## 5 2019-06-18 Jun   Tue              2620 11116              825              106
## 6 2019-06-20 Jun   Thu              2812 12032              954              135
## # … with 11 more variables: Minutes_Fairly_Active <dbl>,
## #   Minutes_Very_Active <dbl>, Activity_Calories <dbl>, Minutes_Asleep <dbl>,
## #   Minutes_Awake <dbl>, Number_Awakenings <dbl>, Time_In_Bed <dbl>,
## #   Minutes_REM_Sleep <int>, Minutes_Light_Sleep <int>,
## #   Minutes_Deep_Sleep <int>, Minutes_Active <dbl>

3. Data Analysis

3.1 Steps

For steps analysis, we will use original data since there are not missing steps.

How many steps did I take in 392 days?

summarise(fitbit_data,Total_Steps = sum(Steps))

## # A tibble: 1 x 1
##   Total_Steps
##         <dbl>
## 1     3657084

3.6 million steps!

On average, I walk or run a mile in about 2,000 steps; therefore total distance traveled by feet in 56 weeks is 1829 miles.

summarise(fitbit_data,Miles = sum(Steps)/2000)

## # A tibble: 1 x 1
##   Miles
##   <dbl>
## 1 1829.

Let’s plot line chart to see distribution of steps.

fitbit_data  %>% ggplot( aes(x=Date, y= Steps)) +
    geom_line( color="red")  + scale_x_date(date_labels = "%b\n%Y", date_breaks = "1 month", limits = c(as.Date("2019-06-01"),as.Date("2020-07-04")))

My daily steps goal is 8,000. I never missed the goal until Jan 26, 2020 - the day I didn’t reach target to honor tragic loss of NBA legend Kobe Bryant, Rest in Peace!

Daily average steps were about 10,000 per day until Feb 2020. What happened March through June?

I am sure you guessed it right. Like most people I was under shelter-in-place due to COVID-19. I started to hit my goal again as restrictions were being lifted in June.

How are steps distributed?

fitbit_data %>% select(Steps) %>% plot_normality()

Steps are approximately normally distributed. It is further illustrated visually by violin plot.

#Violin Plot
ggplot(fitbit_data, aes(x = Date, y = Steps)) + geom_violin(fill = 'green', alpha= 0.5, position = 'dodge') + xlab("Month") + ylab("No. of Steps")

Let’s explore steps weekly and then monthly.

ggplot(fitbit_data, mapping = aes(x= reorder(Day,Steps, FUN = mean ),y = Steps)) + geom_boxplot()

I am impressed with my performance on Mondays - pretty consistent and second best average steps.

# monthly
ggplot(fitbit_data, mapping = aes(x= reorder(Month,Steps, FUN = median ),y = Steps)) + geom_boxplot()

Winter is rainy in the Bay Area; it has negative impact on the movement.

3.2 All Variables

Let’s explore all activities.

p1 <- fitbit_data1 %>% group_by(Day) %>% describe(Minutes_Active) %>% select(variable, Day, mean) %>% ggplot(aes(x=reorder(Day, -mean), y=mean)) + geom_bar(stat="identity", fill="steelblue")+ xlab("Day") + ylab(" Minutes Active") + theme_minimal()

p2<- fitbit_data1 %>% group_by(Day) %>% describe(Minutes_Asleep) %>% select(variable, Day, mean) %>% ggplot(aes(x=reorder(Day, -mean), y=mean)) + geom_bar(stat="identity", fill="steelblue")+ xlab("Day") + ylab(" Minutes Asleep") + theme_minimal()

p3<- fitbit_data1 %>% group_by(Day) %>% describe(Minutes_Sedentary) %>% select(variable, Day, mean) %>% ggplot(aes(x=reorder(Day, -mean), y=mean)) + geom_bar(stat="identity", fill="steelblue")+ xlab("Day") + ylab(" Minutes Sedentary") + theme_minimal()

p4<- fitbit_data1 %>% group_by(Day) %>% describe(Steps) %>% select(variable, Day, mean) %>% ggplot(aes(x=reorder(Day, -mean), y=mean)) + geom_bar(stat="identity", fill="steelblue")+ xlab("Day") + ylab(" Steps") + theme_minimal()

# plot side by side
ggarrange(p1,p2,p3,p4)

Fridays are pretty erratic. I move and sleep more over the weekend.

How do I spend my days?

fitbit_data1 %>% describe(Minutes_Sedentary,Minutes_Active,Time_In_Bed) %>% select(variable, mean) %>% ggplot(aes(x=reorder(variable, mean), y=mean)) + geom_bar(stat="identity", fill="steelblue", width = 0.5)+ xlab("Activities") + ylab(" Minutes ") + theme_minimal()

That’s a lot of sedentary minutes. I would love to bring those minutes down a bit.

Let’s breakdown sleep cycles and calculate minutes spent in each stage.

# pie chart
p5<- fitbit_data1 %>% describe(Minutes_Awake,Minutes_REM_Sleep,Minutes_Light_Sleep,Minutes_Deep_Sleep) %>% select(variable, mean) 
pie <- ggplot(p5, aes(x = "", y=mean, fill = factor(variable)))+ geom_bar(width = 1, stat = "identity") + geom_text(aes(label = paste(round(mean/sum(mean)*100), "%")), position = position_stack(vjust = 0.5)) + theme_classic() + 
 theme(plot.title = element_text(hjust = 0.5),
       axis.line = element_blank(),
       axis.text = element_blank(),
       axis.ticks = element_blank()) +
  labs(fill= "Sleep Stages", x = NULL, y = NULL)
pie + coord_polar("y")

Minutes Awakes are high. It is contributed largely by my reluctance to leave bed when alarm goes off in the morning.

Next, we compute the statistics with all variables numerically and graphically.

fitbit_data1 %>% describe() %>% select(variable, mean,sd, p00,p50, p100)

## # A tibble: 15 x 6
##    variable                 mean      sd   p00    p50  p100
##    <chr>                   <dbl>   <dbl> <dbl>  <dbl> <dbl>
##  1 Calories_Burned        2662.   235.    2046 2651    3661
##  2 Steps                  9462.  2878.    1928 9446.  19525
##  3 Minutes_Sedentary       786.   129.     445  774    1401
##  4 Minutes_Lightly_Active  191.    44.1     39  191     385
##  5 Minutes_Fairly_Active    14.8   11.4      0   13      69
##  6 Minutes_Very_Active      31.6   19.7      0   30     111
##  7 Activity_Calories      1125.   294.     131 1127    2406
##  8 Minutes_Asleep          355.    54.7    152  354.    529
##  9 Minutes_Awake            78.2   17.2     36   77.5   143
## 10 Number_Awakenings        32.9    8.23    13   33      60
## 11 Time_In_Bed             433.    66.6    188  431     654
## 12 Minutes_REM_Sleep        59.9   25.1      4   58     129
## 13 Minutes_Light_Sleep     230.    42.1    106  230     386
## 14 Minutes_Deep_Sleep       65.0   18.5      7   66     114
## 15 Minutes_Active          238.    51.3     39  238     440

p00 is minimum, p50 is median and p100 is max value.

# bar plot with avg of all variables 
fitbit_data1 %>% describe() %>% select(variable, mean) %>% ggplot(aes(x=reorder(variable, mean), y=mean)) + geom_bar(stat="identity", fill="steelblue")+ xlab("Activities") + ylab("Counts ") + theme_minimal() + coord_flip()

3.3 Time Series Plot

Let’s plot variables with interactive time series plot to see pattern over the time.

ft_timeSeries <- xts(x = select(fitbit_data,-(Date:Month)),
                                order.by = fitbit_data$Date)
dygraph(ft_timeSeries, main = "Time Series Plot")

3.4 Correlation

Finally, let’s see if variables are correlated. Note: correlation doesn’t imply causation.

fitbit_data2 <- fitbit_data1 %>% select(-Day,-Month,-Date )
ggcorrplot(cor(fitbit_data2), hc.order = TRUE)

Sleeping more doesn’t necessarily guarantee good REM and deep sleep. Steps are highly correlated with calories burned.

Data Viz