0
$\begingroup$

I was having an idea for a software, which would make use of machine learning, and I started to code it. I got stuck at selecting the algorithm, since I'm not familiar with this field.

My use-case is:

For teaching, I'm collecting Wi-Fi data, and measure signal strength, like:

My network, 86%
Neighbours network, 34%
Other network, 23%

Then I assign 2 numbers to it, each between -1 and +1, so for example: (-0.65, 0.43)

For the evaluation part, I give similar samples:

My network, 68%
Neighbours network, 27%
Other network, 36%

Then expect the algorithm to return something like: (-0.63, 0.41)

I'm thinking it would be neural networks or pattern recognition, but these are really broad terms. I was hoping someone could tell me which algorithm(s) would cope well with this problem.

$\endgroup$
3
  • $\begingroup$ I don't understand the question. How do you go from your 3 percentages to the two numbers between -1 and +1 ? $\endgroup$ – sqrt Nov 7 '15 at 11:15
  • $\begingroup$ By teaching the algorithm. $\endgroup$ – Daniel Zolnai Nov 7 '15 at 12:09
  • $\begingroup$ So you want to predict ground coordinates as function of signal strengths? You could also predict room as function of signal strengths. That might be a little easier for a start. Walls etc. will reduce the signal in a non-linear fashion so a non-linear model would probably be best. Neural-nets with one input for each wifi plus one or two hidden layers and two outputs may work well. You may need 500 training samples. $\endgroup$ – Soren Havelund Welling Nov 7 '15 at 13:07
0
$\begingroup$

I would say that this is not a use case for machine learning. Since each data point has only one dimension you can just intelligibly pick a threshold. Density estimation (KDE, GMM, etc..) would help you determine how to do this correctly.

If you are using this as a simple example of a machine learning algorithm, then try logistic regression. It will project your data between 0 and 1, from which point you can scale it between -1 and 1.

$\endgroup$

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.