Sleep_Disorder
Romario
2023-09-06
Introduction ———————-
Sleep is an important part of human life. When we are sleeping, our brains can retain the knowledge we have acquired throughout the day and restore our energy.
However, nowadays, it is really common for adults not to have a good sleep time, perhaps due to their occupations, daily stress, or other bad habits.
So, can machine learning help them get a better night’s sleep?
Goals
Train a machine learning model capable of predicting whether someone will have a sleep disorder based on their lifestyle.
Exploring Dataset
Libraries ———————-
library(here)
library(tidyverse)Dataset ———————-
data_raw = readr::read_csv(file = here::here('data/raw/Sleep_health_and_lifestyle_dataset.csv'))
data_raw |> head(5)## # A tibble: 5 × 13
## `Person ID` Gender Age Occupation `Sleep Duration` `Quality of Sleep`
## <dbl> <chr> <dbl> <chr> <dbl> <dbl>
## 1 1 Male 27 Software Engineer 6.1 6
## 2 2 Male 28 Doctor 6.2 6
## 3 3 Male 28 Doctor 6.2 6
## 4 4 Male 28 Sales Representa… 5.9 4
## 5 5 Male 28 Sales Representa… 5.9 4
## # ℹ 7 more variables: `Physical Activity Level` <dbl>, `Stress Level` <dbl>,
## # `BMI Category` <chr>, `Blood Pressure` <chr>, `Heart Rate` <dbl>,
## # `Daily Steps` <dbl>, `Sleep Disorder` <chr>Clean Dataset ———————-
Rename Dataset
print('--------------Before Rename-----------------')## [1] "--------------Before Rename-----------------"colnames(data_raw)## [1] "Person ID" "Gender"
## [3] "Age" "Occupation"
## [5] "Sleep Duration" "Quality of Sleep"
## [7] "Physical Activity Level" "Stress Level"
## [9] "BMI Category" "Blood Pressure"
## [11] "Heart Rate" "Daily Steps"
## [13] "Sleep Disorder"print('--------------------After Rename------------------')## [1] "--------------------After Rename------------------"colnames(data_raw)=lapply(colnames(data_raw),
function(x) stringr::str_replace_all(string=x,pattern=" ", repl="_")) |>
tolower()
data_renamed = data_raw
colnames(data_renamed)## [1] "person_id" "gender"
## [3] "age" "occupation"
## [5] "sleep_duration" "quality_of_sleep"
## [7] "physical_activity_level" "stress_level"
## [9] "bmi_category" "blood_pressure"
## [11] "heart_rate" "daily_steps"
## [13] "sleep_disorder"Transform character and integer in factors
transform_in_factor=c("gender","occupation","bmi_category","sleep_disorder",'stress_level','quality_of_sleep')
for(x in transform_in_factor){
data_renamed[x] = factor(data_renamed[[x]])
}
rm(x)
rm(transform_in_factor)
data_renamed_factored = data_renamedRename Factor “Normal Weight” for “Underweight”
data_renamed_factored$bmi_category <- recode(data_renamed_factored$bmi_category,`Normal Weight`='Underweight')summary(data_renamed_factored)## person_id gender age occupation sleep_duration
## Min. : 1.00 Female:185 Min. :27.00 Nurse :73 Min. :5.800
## 1st Qu.: 94.25 Male :189 1st Qu.:35.25 Doctor :71 1st Qu.:6.400
## Median :187.50 Median :43.00 Engineer :63 Median :7.200
## Mean :187.50 Mean :42.18 Lawyer :47 Mean :7.132
## 3rd Qu.:280.75 3rd Qu.:50.00 Teacher :40 3rd Qu.:7.800
## Max. :374.00 Max. :59.00 Accountant:37 Max. :8.500
## (Other) :43
## quality_of_sleep physical_activity_level stress_level bmi_category
## 4: 5 Min. :30.00 3:71 Normal :195
## 5: 7 1st Qu.:45.00 4:70 Underweight: 21
## 6:105 Median :60.00 5:67 Obese : 10
## 7: 77 Mean :59.17 6:46 Overweight :148
## 8:109 3rd Qu.:75.00 7:50
## 9: 71 Max. :90.00 8:70
##
## blood_pressure heart_rate daily_steps sleep_disorder
## Length:374 Min. :65.00 Min. : 3000 Insomnia : 77
## Class :character 1st Qu.:68.00 1st Qu.: 5600 None :219
## Mode :character Median :70.00 Median : 7000 Sleep Apnea: 78
## Mean :70.17 Mean : 6817
## 3rd Qu.:72.00 3rd Qu.: 8000
## Max. :86.00 Max. :10000
## Split blood pressure
data_renamed_factored_new_features= data_renamed_factored |> separate_wider_delim(blood_pressure, "/", names = c("systolic_pressure", "diastolic_pressure"))
for(x in c("systolic_pressure", "diastolic_pressure")) {
data_renamed_factored_new_features[x] = as.numeric(data_renamed_factored_new_features[[x]])
}
rm(x)
data_renamed_factored_new_features |> summary()## person_id gender age occupation sleep_duration
## Min. : 1.00 Female:185 Min. :27.00 Nurse :73 Min. :5.800
## 1st Qu.: 94.25 Male :189 1st Qu.:35.25 Doctor :71 1st Qu.:6.400
## Median :187.50 Median :43.00 Engineer :63 Median :7.200
## Mean :187.50 Mean :42.18 Lawyer :47 Mean :7.132
## 3rd Qu.:280.75 3rd Qu.:50.00 Teacher :40 3rd Qu.:7.800
## Max. :374.00 Max. :59.00 Accountant:37 Max. :8.500
## (Other) :43
## quality_of_sleep physical_activity_level stress_level bmi_category
## 4: 5 Min. :30.00 3:71 Normal :195
## 5: 7 1st Qu.:45.00 4:70 Underweight: 21
## 6:105 Median :60.00 5:67 Obese : 10
## 7: 77 Mean :59.17 6:46 Overweight :148
## 8:109 3rd Qu.:75.00 7:50
## 9: 71 Max. :90.00 8:70
##
## systolic_pressure diastolic_pressure heart_rate daily_steps
## Min. :115.0 Min. :75.00 Min. :65.00 Min. : 3000
## 1st Qu.:125.0 1st Qu.:80.00 1st Qu.:68.00 1st Qu.: 5600
## Median :130.0 Median :85.00 Median :70.00 Median : 7000
## Mean :128.6 Mean :84.65 Mean :70.17 Mean : 6817
## 3rd Qu.:135.0 3rd Qu.:90.00 3rd Qu.:72.00 3rd Qu.: 8000
## Max. :142.0 Max. :95.00 Max. :86.00 Max. :10000
##
## sleep_disorder
## Insomnia : 77
## None :219
## Sleep Apnea: 78
##
##
##
## EDA ———————-
data_eda = data_renamed_factored_new_features
data_eda = data_eda |> dplyr::select(-c('person_id'))Numeric Features
columns_numeric = data_eda |> dplyr::select_if(is.numeric) |> colnames()
#columns_numeric
for( x in columns_numeric){
g = data_eda |> ggplot() +
geom_histogram(aes(.data[[x]]), bins = 30) +
xlab(x)
print(g)
}
rm(columns_numeric,x,g)Sleep Disorder For Categorical Features
columns_factor = data_eda |> dplyr::select_if(is.factor) |> dplyr:: select(-c(sleep_disorder)) |> colnames()
for( x in columns_factor){
g = data_eda |>
ggplot() +
geom_bar(aes(x=sleep_disorder,
fill=.data[[x]]),
position ='dodge2')
print(g)
}
rm(columns_factor,x,g)Train Models ———————-
Libraries
library(here)
library(tidymodels)
library(tidyverse)
library(rpart.plot)Split Dataset
set.seed(1)
split <- rsample::initial_split(data_eda, prop = .80,strata = sleep_disorder)
train_set <- rsample::training(split)
test_set <- rsample::testing(split)Decision Tree
model_decision_tree=parsnip::decision_tree(mode='classification', engine = 'rpart')
model_decision_tree_fit= model_decision_tree |> fit( train_set$sleep_disorder ~ ., data = train_set)
graph_model_decision_tree_fit <- extract_fit_engine(model_decision_tree_fit)
rpart.plot::rpart.plot(graph_model_decision_tree_fit)
SVM Linear
model_svm_linear=parsnip::svm_linear(mode ='classification',engine = 'kernlab')
model_svm_linear_fit= model_svm_linear |> fit( train_set$sleep_disorder ~ ., data = train_set) ## Setting default kernel parametersmodel_svm_linear_fit## parsnip model object
##
## Support Vector Machine object of class "ksvm"
##
## SV type: C-svc (classification)
## parameter : cost C = 1
##
## Linear (vanilla) kernel function.
##
## Number of Support Vectors : 89
##
## Objective Function Value : -23.4147 -18.1426 -25.4937
## Training error : 0.077181
## Probability model included.SVM Radial
model_svm_radial=parsnip::svm_rbf(mode ='classification',engine = 'kernlab')
model_svm_radial_fit= model_svm_radial |> fit( train_set$sleep_disorder ~ ., data = train_set)
model_svm_radial_fit## parsnip model object
##
## Support Vector Machine object of class "ksvm"
##
## SV type: C-svc (classification)
## parameter : cost C = 1
##
## Gaussian Radial Basis kernel function.
## Hyperparameter : sigma = 0.0930375075840404
##
## Number of Support Vectors : 93
##
## Objective Function Value : -31.1503 -25.7571 -31.9414
## Training error : 0.087248
## Probability model included.Random Forest
model_random_forest= rand_forest() |>
set_engine('randomForest') |>
set_mode('classification')
model_random_forest_workflow = workflow() |>
add_model(model_random_forest) |>
add_formula(sleep_disorder ~ .)
model_random_forest_fit = fit(model_random_forest_workflow, train_set)Evaluate Model ———————-
Preprocess
test_set_without_class=subset(test_set,select = -c(sleep_disorder))
test_set_only_class=subset(test_set,select = c(sleep_disorder))
classification_metrics <- yardstick::metric_set(accuracy, mcc, f_meas)Decision Tree
predict(model_decision_tree_fit,test_set_without_class ) |>
dplyr::bind_cols(test_set_only_class) |>
yardstick::conf_mat(truth = sleep_disorder, estimate = .pred_class)## Truth
## Prediction Insomnia None Sleep Apnea
## Insomnia 10 0 0
## None 6 44 1
## Sleep Apnea 0 0 15predict(model_decision_tree_fit,test_set_without_class ) |>
dplyr::bind_cols(test_set_only_class) |>
classification_metrics(truth = sleep_disorder, estimate = .pred_class)## # A tibble: 3 × 3
## .metric .estimator .estimate
## <chr> <chr> <dbl>
## 1 accuracy multiclass 0.908
## 2 mcc multiclass 0.844
## 3 f_meas macro 0.888SVM Linear
predict(model_svm_linear_fit,test_set_without_class ) |>
dplyr::bind_cols(test_set_only_class) |>
yardstick::conf_mat(truth = sleep_disorder, estimate = .pred_class)## Truth
## Prediction Insomnia None Sleep Apnea
## Insomnia 13 1 0
## None 2 43 0
## Sleep Apnea 1 0 16predict(model_svm_linear_fit,test_set_without_class ) |>
dplyr::bind_cols(test_set_only_class) |>
classification_metrics(truth = sleep_disorder, estimate = .pred_class)## # A tibble: 3 × 3
## .metric .estimator .estimate
## <chr> <chr> <dbl>
## 1 accuracy multiclass 0.947
## 2 mcc multiclass 0.909
## 3 f_meas macro 0.934SVM Radial
predict(model_svm_radial_fit,test_set_without_class ) |>
dplyr::bind_cols(test_set_only_class) |>
yardstick::conf_mat(truth = sleep_disorder, estimate = .pred_class)## Truth
## Prediction Insomnia None Sleep Apnea
## Insomnia 11 0 0
## None 2 44 0
## Sleep Apnea 3 0 16predict(model_svm_radial_fit,test_set_without_class ) |>
dplyr::bind_cols(test_set_only_class) |>
classification_metrics(truth = sleep_disorder, estimate = .pred_class)## # A tibble: 3 × 3
## .metric .estimator .estimate
## <chr> <chr> <dbl>
## 1 accuracy multiclass 0.934
## 2 mcc multiclass 0.889
## 3 f_meas macro 0.902Random Florest
predict(model_random_forest_fit,test_set_without_class ) |>
dplyr::bind_cols(test_set_only_class) |>
yardstick::conf_mat(truth = sleep_disorder, estimate = .pred_class)## Truth
## Prediction Insomnia None Sleep Apnea
## Insomnia 11 0 0
## None 2 44 0
## Sleep Apnea 3 0 16predict(model_random_forest_fit,test_set_without_class ) |>
dplyr::bind_cols(test_set_only_class) |>
classification_metrics(truth = sleep_disorder, estimate = .pred_class)## # A tibble: 3 × 3
## .metric .estimator .estimate
## <chr> <chr> <dbl>
## 1 accuracy multiclass 0.934
## 2 mcc multiclass 0.889
## 3 f_meas macro 0.902Conclusion
The model with the best metrics was the SVM with Linear Kernel. However, its performance can become better with some tuning techniques.