0

So a wearable device like Fitbit has data from accelerometer( movements in the wearable(watch) ) and also heart rate depending upon the device. And using the data from these inputs, fitbit is able to predict how much user has slept. So data I have looks like :

Time              |Screen | Mobile_state | Plugged/Unplugged| 

10PM    - 10:30PM | ON    |  Moving      |  Unplugged       
11PM    - 11:15PM | ON    |  Moving      |  Plugged         
11:15PM - 11:51PM | OFF   |  Idle        |  Plugged         
11:51PM - 6:00AM  | OFF   |  Idle        |  Plugged         
6AM     - 6:10AM  | ON    |  Idle        |  UnPlugged       
6:10AM  - 8:00AM  | OFF   |  Idle        |  UnPlugged

So here I have 3 parameters, using which i can say user slept from 11:15 till 6AM.
In reality i may have 5-6 params like these. What kind of ML algorithm might be useful here which takes this kind of input and gives (11:15PM,6AM) as the output

pradeep bansal
  • 26
  • 1
  • 5
  • Why use machine learning. Not moving and heart rate should for more than lets say 1h should be a quite good indicator for being asleep. Likely times when 'normal' people sleep (= at night) should even improve this prediction. – MrSmith42 Jun 20 '17 at 11:29
  • How does it differentiate between simply idle and charging vs sleeping? Are you wearing it to sleep? It tracks heart rate too, right? – OneCricketeer Jun 20 '17 at 11:30
  • @cricket_007 There's a strong assumption, that if mobile is idle in night for large duration, user must be on bed and sleeping. Hence calling it sleep/bed time. It doesn't track heart rate. – pradeep bansal Jun 21 '17 at 13:03
  • @MrSmith42 I am just thinking if we could do it using ML. We don't have heart rate here as this is just a normal smart phone like Android or iPhone. Right now I have written 3 parameters here (Screen, Mobile_state, Plugged). In real, we will have few more and making rules based on all of parameters could be difficult. – pradeep bansal Jun 21 '17 at 13:04

0 Answers0