Analyzing my Oura sleep score - is it AI or just math?

Today, I delved into my Oura ring's sleep score data, prompting a crucial question: Does this problem require AI, or will a simple formula suffice?

Activity Tracking and the Oura Ring

I'm a data-driven health enthusiast, utilizing fitness trackers like Fitbit and Garmin. However, for everyday wear, I prefer the Oura ring for its discreet design. Sleep tracking is a key feature, making its sleep score worthy of investigation. (For those unfamiliar, Oura's sleep score is detailed on their blog.)

The Enigmatic Sleep Score

A drawback of Oura is its paywalled insights. The free version only displays the sleep score, unlike Fitbit and Garmin's comprehensive dashboards. This raises the question: What makes this sleep score so special, and is the subscription worth it?

The Hypothesis: Simple Correlations

My initial hypothesis, as a data scientist, was straightforward: Higher deep sleep duration and lower average heart rate correlate with better sleep scores. Could it be that simple? Let's find out.

Data Acquisition and Processing

I accessed my Oura data via their developer API, retrieving sleep data and saving it as a JSON file.

def get_data(type):
  url = 'https://api.ouraring.com/v2/usercollection/' + type
  params={ 
      'start_date': '2021-11-01', 
      'end_date': '2025-01-01' 
  }
  headers = { 
    'Authorization': 'Bearer ' + auth_token 
  }
  response = requests.request('GET', url, headers=headers, params=params) 
  return response.json()["data"]

data = get_data("sleep")
with open('oura_data_sleep.json', 'w', encoding='utf-8') as f:
    json.dump(data, f, ensure_ascii=False, indent=4)

This data was then indexed in Elasticsearch for easy querying. The JSON structure simplified this process, requiring no extra mapping or data cleaning.

client = Elasticsearch(
    cloud_id=ELASTIC_CLOUD_ID,  
    api_key=ELASTIC_API_KEY 
)

index_name = 'oura-history-sleep'

# ... (Elasticsearch index creation and data loading code) ...

The Experiment: Simple Queries

My experiment involved simple queries to test my hypothesis. I first sorted days by highest sleep score:

response = client.search(index = index_name, sort="readiness.score:desc")

# ... (Code to print day and sleep score) ...

Examining these high-scoring days revealed consistent patterns in deep sleep and heart rate. Then, I built an Elasticsearch query filtering for deep sleep over 1.5 hours and heart rate under 60 bpm, sorted by REM sleep:

query = {
    "range" : {
        "deep_sleep_duration" : {
            "gte" : 1.5*3600
        }
    },
    "range" : {
        "average_heart_rate":{
            "lte" : 60
        }
    }
}
response = client.search(index = index_name, query=query, sort="rem_sleep_duration:desc")

The results strongly correlated with the initial high-score days. While not perfect, this demonstrates the predictive power of a simple formula. Further Kibana visualizations (shown below) reinforce this connection.

The Significance

In the hype surrounding AI, it's easy to overlook simpler solutions. This sleep score, often presented as a complex AI achievement, is essentially based on a straightforward formula. This highlights the importance of understanding when simpler methods are sufficient – leading to more accurate, cost-effective, and easily interpretable results. This underscores the enduring value of data science fundamentals and intuitive modeling. While advanced technology is impressive, knowing when not to use it is equally crucial.

See full code notebook here.

The above is the detailed content of Analyzing my Oura sleep score - is it AI or just math?. For more information, please follow other related articles on the PHP Chinese website!