Ideas for a classifica​tion/regre​ssion problem.

Hi,

I have some data, like:

X: Car_Weigth, Car_type, Weather_type, Speed_Before_Accident, Drunk_Pilot_Status Y: 1, 2, 3, 4........ 100 miles.

For me, this is a classification problem, easily solved by any classification method, like: 5 inputs for a neural and 100 outputs.

The doubt here, is that, I have 100 classes, and so, what kind of technique can i use to solve this kind of problem ?

any idea would be great.

[Ilya]

This sounds like a regression problem to me. Why do you prefer classification?

[Andre]

Hi Ilya!!

sounds great for me, and I'm trying to solve this not by a classification problem (imagina if I have 1000miles - this would get me 1000 classes).

But, I don't know how to start. For me (wrong I think), regression problems just solve temporal prediction problems...

Can you please give me an idea ??

[Greg Heath]

I don't understand. Please explain and give an example of what you call a class.

fitnet: regression or curve-fitting; NOT temporal prediction

patternnet: classification or pattern-recognition

 For the latter case with c classes, the targets should be unit vectors with c-1 zeros and a single one.

timedlaynet, narnet and narxnet: temporal prediction

 Use the help and doc commands for details of any function.

Answer 1

[Ilya]

You could start with linear regression. If you have the Statistics Toolbox, take a look at LinearModel. Or, if you have an old version of the toolbox, take a look at the regress function. If you don't have the Statistics Toolbox, try the backslash operator \.

For non-parametric regression, you could use decision trees, TreeBagger or fitensemble, all in the Statistics Toolbox. You could use the Neural Network toolbox too.

Answer 2

[Andre]

Thank you very much.