0
$\begingroup$

I want to predict which device got used in which room. Therefore I've got device and sensor data.

My idea was to create a feature vector lie this:

             ----------------------------------------------------------
Data-Vector: | u_1 u_2 u_3 | x_1 ... x_7 | y_1 ... y_12 | z_1 ... z_4 |
             ----------------------------------------------------------
 Categories: | device_data | room 1 data | room 2 data  | room 3 data |
             ----------------------------------------------------------

My device data contains amongst other things:
+ timestamps when the device got turned on/off
+ average power consumption and divergences

My room data contains for example:
+ sensor data of motion detector and timestamps
+ sensor data of lamps (turned on/off) and timestamps
+ weather data

In the feature vector I've got the room data closest to the turn on/off timestamp.

All data points itself are floats.

My idea was to use k-means for clustering.
My problems are:
1. When using k-means, how can I tell which cluster correlates to which room label (room1, room2 or room3)?
2. I think it could be beneficial if I add somehow the information: which sensor is which room.

Can I manipulate the data in k-means algorithm so it will only consider:
the device data and room 1 data for the first cluster and sets everything else to zero
the device data and room 2 data for the second cluster
and so on...

This way I could tell that cluster x correlates to room x.

Or will this somehow break the k-means algorithm?

$\endgroup$
0
$\begingroup$

If you have apriori known labels, then use the labels, not clustering.

Clustering is quite difficulty and fragile, so always prefer a simpler solution if you can.

It's not very clear what you are trying to solve. Select columns to split the data into rooms? Just select the columns then.

|improve this answer|||||
$\endgroup$
  • $\begingroup$ I have the power consumption of devices and some sensor data of rooms. I know which sensors are in which room, but I don't know anything about the devices. My goal is to predict which device got turned on in which room based on the sensor and device data. Since I don't know anything about the devices or the correlation of the sensor data, I can not label the vector and thus need clustering, right? $\endgroup$ – GAV3 Jun 12 '18 at 13:45
  • $\begingroup$ What is the device data then? We don't have your data. We can only guess! You need to ask much more precise questions. $\endgroup$ – Has QUIT--Anony-Mousse Jun 12 '18 at 19:51
  • $\begingroup$ Ok I added to my question above, what my data contains and what exactly my problems and ideas to fix them are. $\endgroup$ – GAV3 Jun 13 '18 at 14:37

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.