Apple Health Data - Workout Activity

health
analysis
R
Author

Matthew Harris

Published

January 10, 2021

Abstract
Analyzing my recent run activity.

Packages

Code 
library(dplyr)
library(tidyr)
library(ggplot2)
library(xml2)
library(lubridate)
library(janitor)
library(scales)
library(ggsci)

XML to RDS

The XML file format is great for storing data, but I will want to extract and transform the information that I want into separate data frames. This can be accomplished easily by creating a function that extracts all nodes that contain a given attribute.

Code 
xml_to_rds <- function(xml_file, attr_select, out_path) {
  
  xml2::xml_find_all(xml_file, 
                     paste0("//", attr_select)) %>% 
    purrr::map(~xml2::xml_attrs(.x)) %>%
    purrr::map_df(~as.list(.x)) %>% 
    saveRDS(file = paste0(out_path, "/", attr_select, ".RDS"))
}

The top of the export.xml file contains some documentation on what elements are present within the file. I’ll focus primarily on the Workout nodes for this post.

Code 
apple_health_export <- read_xml(x = "local_data/export.xml")
xml_to_rds(xml_file = apple_health_export, attr_select = "Workout", out_path = "local_data")

Data Wrangling

Great. So now I have data frame that contains all of my workout data. Before I can begin any type of analysis, I’ll need to perform a couple cleaning operations. I’ll start by cleaning up the column names.

Code 
names(workouts)
 [1] "workoutActivityType"   "duration"              "durationUnit"         
 [4] "totalDistance"         "totalDistanceUnit"     "totalEnergyBurned"    
 [7] "totalEnergyBurnedUnit" "sourceName"            "sourceVersion"        
[10] "creationDate"          "startDate"             "endDate"              
[13] "device"
Code 
workouts <- workouts %>% 
  clean_names()

names(workouts)
 [1] "workout_activity_type"    "duration"                
 [3] "duration_unit"            "total_distance"          
 [5] "total_distance_unit"      "total_energy_burned"     
 [7] "total_energy_burned_unit" "source_name"             
 [9] "source_version"           "creation_date"           
[11] "start_date"               "end_date"                
[13] "device"

Next I want to make sure that each variable has the correct data type. I can inspect the current column type with glimpse().

Code 
workouts %>% 
  select(-device) %>% 
  glimpse()
Rows: 159
Columns: 12
$ workout_activity_type    <chr> "HKWorkoutActivityTypeRunning", "HKWorkoutAct…
$ duration                 <chr> "30.19625363349915", "31.56661983331044", "38…
$ duration_unit            <chr> "min", "min", "min", "min", "min", "min", "mi…
$ total_distance           <chr> "3.02360045806866", "3.07297265789395", "3.13…
$ total_distance_unit      <chr> "mi", "mi", "mi", "mi", "mi", "mi", "mi", "mi…
$ total_energy_burned      <chr> "296.1570000000004", "294.944", "347.71600000…
$ total_energy_burned_unit <chr> "Cal", "Cal", "Cal", "Cal", "Cal", "Cal", "Ca…
$ source_name              <chr> "Matthew’s Apple Watch", "Matthew’s Apple Wat…
$ source_version           <chr> "3.1", "3.2.2", "3.2.2", "3.2.2", "3.2.2", "3…
$ creation_date            <chr> "2017-05-15 20:13:44 -0400", "2017-05-29 10:3…
$ start_date               <chr> "2017-05-15 19:43:28 -0400", "2017-05-29 10:0…
$ end_date                 <chr> "2017-05-15 20:13:40 -0400", "2017-05-29 10:3…

It looks as though all of the variables are currently categorized as characters. I can fix this using a combination of the mutate() and across() functions. I’m also adjusting the time to match my current timezone. Most of the data that I will be analyzing was captured during 2020, so I’m pretty sure I didn’t travel to any other time-zones.

Code 
workouts_clean <- workouts %>% 
  select(-device) %>% 
  mutate(across(.cols = contains("date"),
                .fns = ~as_datetime(.x) - seconds(18000)),
         across(.cols = c(duration, total_distance, 
                          total_energy_burned),
                .fns = ~as.double(.x)))

workouts_clean %>% 
  glimpse()
Rows: 159
Columns: 12
$ workout_activity_type    <chr> "HKWorkoutActivityTypeRunning", "HKWorkoutAct…
$ duration                 <dbl> 30.19625, 31.56662, 38.16131, 30.21184, 31.31…
$ duration_unit            <chr> "min", "min", "min", "min", "min", "min", "mi…
$ total_distance           <dbl> 3.023600, 3.072973, 3.137175, 3.098805, 2.941…
$ total_distance_unit      <chr> "mi", "mi", "mi", "mi", "mi", "mi", "mi", "mi…
$ total_energy_burned      <dbl> 296.1570, 294.9440, 347.7160, 287.5090, 272.5…
$ total_energy_burned_unit <chr> "Cal", "Cal", "Cal", "Cal", "Cal", "Cal", "Ca…
$ source_name              <chr> "Matthew’s Apple Watch", "Matthew’s Apple Wat…
$ source_version           <chr> "3.1", "3.2.2", "3.2.2", "3.2.2", "3.2.2", "3…
$ creation_date            <dttm> 2017-05-15 19:13:44, 2017-05-29 09:37:41, 20…
$ start_date               <dttm> 2017-05-15 18:43:28, 2017-05-29 09:06:03, 20…
$ end_date                 <dttm> 2017-05-15 19:13:40, 2017-05-29 09:37:37, 20…

Analysis and Visualization

On to the fun stuff. With my cleaned and formatted data, I can finally start to answer some questions. Let’s determine what workout activity I’ve done the most, on which days, and at what times.

Code 
workouts_clean %>% 
  count(workout_activity_type)
# A tibble: 2 × 2
  workout_activity_type            n
  <chr>                        <int>
1 HKWorkoutActivityTypeRunning   138
2 HKWorkoutActivityTypeWalking    21
Code 
workouts_clean <- workouts_clean %>% 
  filter(workout_activity_type == "HKWorkoutActivityTypeRunning") %>% 
  mutate(workout_day = weekdays(start_date),
         workout_day = factor(workout_day, 
                              levels = c("Monday", "Tuesday",
                                         "Wednesday", "Thursday",
                                         "Friday", "Saturday",
                                         "Sunday")),
         workout_hour = hour(start_date))

workouts_clean %>% 
  ggplot(aes(workout_day)) + geom_bar(fill = "#4789CC") +
  scale_y_continuous(breaks = breaks_width(5)) +
  labs(x = "Weekday", y = "Count") +
  theme_minimal(base_size = 14)

Code 
workouts_clean %>% 
  ggplot(aes(workout_hour)) + geom_bar(fill = "#4789CC") +
  scale_x_continuous(breaks = breaks_width(2)) +
  labs(x = "Hour", y = "Count") +
  theme_minimal(base_size = 14)

So far I have 138 runs recorded. I’ve tried to stick to a running schedule of Monday, Wednesday, Friday, and Sunday. I also tend to go on most of my runs before 12:00 PM.

I’ve tried to use running as a good way to stay active during the pandemic and haven’t really focused on increasing my performance. It would still be interesting to see if I have either gotten faster or increased the distance that I’m running.

Code 
workouts_clean %>% 
  filter(workout_activity_type == "HKWorkoutActivityTypeRunning") %>% 
  mutate(min_per_mile = duration(duration / total_distance, "minute"),
         creation_date = date(creation_date)) %>% 
  filter(creation_date > "2020-01-01", 
         total_distance >= 1.5) %>% 
  ggplot(aes(creation_date, 
             min_per_mile, 
             col = total_distance)) + 
  scale_color_gradient(low = "blue", high = "red", 
                       breaks = seq(1.5, 4, 0.5)) +
  geom_point(size = 3.5, alpha = 0.8) +
  scale_y_time(labels = time_format("%M:%S"),
               breaks = breaks_width(30)) +
  scale_x_date(breaks = breaks_width("2 month"), 
               date_labels = "%b %Y") +
  labs(x = "Date", y = "Avg. Minutes / Mile",
       col = "Miles") +
  geom_smooth(method = "lm", se = FALSE, col = "black",
              linetype = 2) +
  theme_minimal(base_size = 14)
`geom_smooth()` using formula 'y ~ x'