Data Analysis
Tizen smartwatches and smartphones connect via SASocket. In this project, smartwatch is the host. Therefore, the smart watch automatically tries to connect to the smartphone.
The data collected is as follows:
- Heart Rate: Once every 60 seconds, measured for 30 seconds.
- Gyroscope Sensor: Similar to acceleration
- Gyroscope Rotation Vector Sensor: smartwatch angle
- Light Sensor: smartwatch light
The data sent to the smartphone to String (JSON) format.
{
"heartRate":64,
"light": 234,
"motion":{
"gyroscope":{
"x":0,
"y":0,
"z":0
},
"gyroscopeRotation":{
"x":0,
"y":0,
"z":0
}
}
}We classify this data using machine learning. However, before using machine learning, we checked the distribution and characteristics of the data.
This is the closest column for drinking. Drinking alcohol usually improves your heart rate.
The X axis is seconds and the Y axis is heart rate.
The 'red point' is the outlier data, and the 'purple star' means 10-minute heart rate.
However, heart rate alone cannot distinguish between exercise and alcohol. Both have increasing patterns.
'Light data' means illuminance.
In the environment we investigated, drinking mostly occurred in dark environments. However, the workout generally took place in a bright environment.
This is one point that distinguishes drinking from exercise.
When moving the arm wearing the smart watch, the acceleration sensor changes.
The important point here is that the right-handed person wears a smartwatch on his left arm.
This seems to make the smartphone's acceleration sensor meaningless, but it doesn't.
When you do something like eating or drinking, the acceleration hardly changes because it's hard to keep moving your left hand when eating for 30-40 minutes. However, in the case of exercise, both hands are moved.
Therefore, this is also the main point that distinguishes exercise from drinking with light sensors.
acceleration_value = sqrt(x*x + y*y + z*z)