Accord.NET machine learning on table data to predict next row

819 views Asked by At

Ok, I have these huge data sets I want to analyze, and most of the column values relate to one or more values in other columns. Here is a simple example, in reality there are many more columns:

    t     c1     c2     c3     c4
    0   21.0    2.0   54.2    0.1
   15   22.2    2.2   51.5    0.2
   30   24.4    2.9   52.3    0.4
   45   19.1    3.2   58.7    0.7
   60   18.6    3.4   61.3    0.6
   75   ??.?    ?.?   ??.?    ?.?
   90   ??.?    ?.?   ??.?    ?.?

To simplify, I have logged values with equal time intervals for a long time, and now I want to predict the values for c1, c2, c3 and c4 at t=75 and t=90.

If I make arrays out of these column values, eg:

float[] c1Array = { 21.0, 22.2, 24.4, 19.1, 18.6 };
float[] c2Array = { 2.0, 2.2, 2.9, 3.2, 3.4};
float[] c3Array = { 54.2,........61.3}

and so on...

eg.
c1Array[2] would inter-correspond with c2Array[2], c3Array[2] etc...

How would I utilize Accord.NET machine learning, and what are my options? I've read a bit upon using DecisionTree and ID3Learning - and it looks like that can solve my problem:

A simple example with just x and y values (I need an example with x, y, z,....)

But then I came over this, which is Hidden Markov Models in Accord.NET, and it seems like that might be a better fit.

I am a programmer, not a mathematician, and I find most examples tend to operate with single data sets that would have 1 or two "columns", I need to work on multiple.

I could work with something like the x and y values example I linked above, but with using more columns.

But if there are better functions in my case, I would love to see some examples.

0

There are 0 answers