Using K-Means for Better understanding of Calorie Burning

Introduction to K-Means

K-means clustering is an unsupervised learning algorithm used for data clustering, which groups unlabeled data points into groups or clusters. https://www.ibm.com/think/topics/k-means-clustering

K-Means is one of the most popular clustering techniques in unsupervised learning. The k-means approach uses a mathematical distance metric to group data points into clusters. By switching between allocating data points to clusters based on the existing centroids and selecting centroids based on the current assignment of data points to clusters, K-Means determines the optimal centroids.

Data

The data used for the research is about exercise routines, physical attributes, and fitness metrics. It contains key performance indicators such as heart rate, calories burned, and workout duration. Each entry also included demographic data and experience levels, allowing for a comprehensive analysis of fitness patterns, athlete progression, and health trends.

Data: https://www.kaggle.com/datasets/valakhorasani/gym-members-exercise-dataset

Libraries:

library(dplyr)
library(tidymodels)
library(tidyclust)
library(tidyverse)
library(tidyquant)
library(plotly)
library(stats)
library(ggplot2)
library(factoextra)

• 1)We should start with loading the data, then create a subset of first 100 rows.

full_data <- read_csv("gym.csv")
data <- full_data[1:100, ]
data %>% glimpse()

## Rows: 100
## Columns: 15
## $ Age                             <dbl> 56, 46, 32, 25, 38, 56, 36, 40, 28, 28…
## $ Gender                          <chr> "Male", "Female", "Female", "Male", "M…
## $ `Weight (kg)`                   <dbl> 88.3, 74.9, 68.1, 53.2, 46.1, 58.0, 70…
## $ `Height (m)`                    <dbl> 1.71, 1.53, 1.66, 1.70, 1.79, 1.68, 1.…
## $ Max_BPM                         <dbl> 180, 179, 167, 190, 188, 168, 174, 189…
## $ Avg_BPM                         <dbl> 157, 151, 122, 164, 158, 156, 169, 141…
## $ Resting_BPM                     <dbl> 60, 66, 54, 56, 68, 74, 73, 64, 52, 64…
## $ `Session_Duration (hours)`      <dbl> 1.69, 1.30, 1.11, 0.59, 0.64, 1.59, 1.…
## $ Calories_Burned                 <dbl> 1313, 883, 677, 532, 556, 1116, 1385, …
## $ Workout_Type                    <chr> "Yoga", "HIIT", "Cardio", "Strength", …
## $ Fat_Percentage                  <dbl> 12.6, 33.9, 33.4, 28.8, 29.2, 15.5, 21…
## $ `Water_Intake (liters)`         <dbl> 3.5, 2.1, 2.3, 2.1, 2.8, 2.7, 2.3, 1.9…
## $ `Workout_Frequency (days/week)` <dbl> 4, 4, 4, 3, 3, 5, 3, 3, 4, 3, 2, 3, 3,…
## $ Experience_Level                <dbl> 3, 2, 2, 1, 1, 3, 2, 2, 2, 1, 1, 2, 2,…
## $ BMI                             <dbl> 30.20, 32.00, 24.71, 18.41, 14.39, 20.…

• 2)As we want to find what affects the calorie burning, we can clean the data from columns which will not affect the calorie burning.

data_prep <- data %>%
  drop_na() %>%
  select(-`Water_Intake (liters)`, -Fat_Percentage, -`Weight (kg)`, -`Height (m)`, -Experience_Level, -Max_BPM, 
         - Workout_Type)

• 3)Then we should create separate data frames, which will be helpful in determining the best correlation. We will use session duration and average beats per minute (BPM) columns as they are logically correlated with calorie burning.

cal_bpm = data_prep %>%
  select(Calories_Burned, Avg_BPM)

cal_duration = data_prep %>%
  select(Calories_Burned, `Session_Duration (hours)`)

• 4)Elbow method should be used for finding optimal number of clusters.

fviz_nbclust(cal_bpm, kmeans, method = "wss") +
  labs(subtitle = "Elbow method")

fviz_nbclust(cal_duration, kmeans, method = "wss") +
  labs(subtitle = "Elbow method")

As can be seen from the tables for both data frames, 3 can be selected as the optimal number of clusters.

• 5)Now we can perform K-Means Clustering. By including workout type and index we can make it more convinient:
  • I - Clustering Based on Calories & BPM

km <- kmeans(cal_bpm, centers = 3)

km.cluster <- km$cluster
rownames(cal_bpm) <- paste(data$Workout_Type, 1:dim(data)[1], sep = "-")
fviz_cluster(list(data = cal_bpm, cluster = km.cluster))

• II - Clustering Based on Calories & Duration

km <- kmeans(cal_duration, centers = 3)

km.cluster <- km$cluster
rownames(cal_duration) <- paste(data$Workout_Type, 1:dim(data)[1], sep = "-")
fviz_cluster(list(data = cal_duration, cluster = km.cluster))

The plot indicates that session duration exhibits a stronger correlation with burnt calories, and it can be anlyzed more with interactive data.

• 6)We can utilize tidymodels library for preprocessing, and all numerical data will be standardized using the step_scale(all_numeric_predictors()) function.

recipe <- recipe(~ ., data = cal_duration) %>%
  step_scale(all_numeric_predictors())

recipe %>% prep() %>% juice() %>% glimpse()

## Rows: 100
## Columns: 2
## $ Calories_Burned            <dbl> 4.893835, 3.291132, 2.523325, 1.982879, 2.0…
## $ `Session_Duration (hours)` <dbl> 5.246849, 4.036038, 3.446155, 1.831740, 1.9…

• 7)As next step, we build a K-Means Clustering Model

model_kmeans <- k_means(num_clusters = 3) %>%
  set_engine("stats")
set.seed(123)
calories <- workflow() %>%
  add_model(model_kmeans) %>%
  add_recipe(recipe) %>%
  fit(cal_duration)

• 8)We can use the trained K-Means model to make predictions and analyze the clustering results.

calories %>% predict(cal_duration)

## # A tibble: 100 × 1
##    .pred_cluster
##    <fct>        
##  1 Cluster_1    
##  2 Cluster_2    
##  3 Cluster_3    
##  4 Cluster_3    
##  5 Cluster_3    
##  6 Cluster_1    
##  7 Cluster_1    
##  8 Cluster_2    
##  9 Cluster_3    
## 10 Cluster_3    
## # ℹ 90 more rows

extract_cluster_assignment(calories)

## # A tibble: 100 × 1
##    .cluster 
##    <fct>    
##  1 Cluster_1
##  2 Cluster_2
##  3 Cluster_3
##  4 Cluster_3
##  5 Cluster_3
##  6 Cluster_1
##  7 Cluster_1
##  8 Cluster_2
##  9 Cluster_3
## 10 Cluster_3
## # ℹ 90 more rows

extract_centroids(calories)

## # A tibble: 3 × 3
##   .cluster  Calories_Burned `Session_Duration (hours)`
##   <fct>               <dbl>                      <dbl>
## 1 Cluster_1            4.84                       5.37
## 2 Cluster_2            3.66                       4.05
## 3 Cluster_3            2.43                       2.91

g <- data_prep %>%
  bind_cols(extract_cluster_assignment(calories), .) %>%
  ggplot(aes(Calories_Burned, `Session_Duration (hours)`)) +
  geom_point(
    aes(fill = .cluster),
    shape = 21, alpha = 0.4, size = 5) +
  geom_smooth(color = "red", se = FALSE) +
  scale_x_continuous(labels = scales::number_format(), 
                     limits = c(0, 2000))+
  scale_y_continuous(labels = scales::number_format(),
                     limits = c(0, 3)) +
  labs(title = "Calories burned by Duration", x = "Calories Burnt", y = "Duration of Session")+
  scale_fill_tq()+
  theme_tq()
ggplotly(g)

## `geom_smooth()` using method = 'loess' and formula = 'y ~ x'

Conclusion

This study employed K-Means clustering on exercise routine data to identify patterns in calories burned, average BPM, and session duration. The Elbow Method was utilized to determine the optimal number of clusters, which was found to be three for calories burned versus BPM and session duration. Through the application of clustering techniques, correlations were also identified, and predictions were generated.