Real time classification problem

I have data of two different classes (attached xls file) blue is belongs to one class and red is belongs to another class. I need to classify the future real time inputs based on this data. Please suggest me simple classification technique which can be implemented easily in a microcontroller to do this classification task.

Answer 1

[Neil Caithness]

You could try building a neural net classifier, but there are lots of variations to try and you need to work at it.

Here is a quick start based on your example but you'll need to train it with a much bigger dataset.

xtrain = [
   28.1800  100.3800    3.9900   74.4600  371.3400  119.7200   53.7200   93.3400   18.6000
   24.0800   24.6900   12.9600  241.5400   46.6600  102.4900   81.8700    5.8200   29.4800
  127.6300   19.2300   12.9300    4.0100    6.6700   87.2900  211.8200   50.4200   39.1500
   28.7800  117.4000    0.5800  364.8400    8.8000   76.7600    9.3900   77.2200   18.7300
   24.4700   70.4800   13.4500   56.0200  101.9400   67.0800    1.9500   48.0800   25.7500
  170.8900   20.0300   11.1600  518.9100  620.4600   53.3400   12.0500   46.1600   31.1200
   28.1800  188.9000    0.9600   23.4300   66.4900   47.4700   17.2500   77.6800   11.0700
   25.8400  106.4900   14.0100  335.3600    0.1300   35.2400   21.6200   90.4800   18.2700
  124.9800   17.4800    0.1300   25.2900    1.7700   28.3300   24.5100   87.3500   26.3400
   28.8200  180.4300   11.6900  456.7700   72.5900   25.3300   25.8800    5.8600    2.6700
   26.4700  137.1700    7.5500   53.8800  585.5100   14.6500   24.3900   71.1500   14.6500
   85.8000   24.5400    4.1800  323.0400   81.7500    2.9200   25.8800   48.8500   20.9100
   29.2000   18.8800   13.2600    6.5300    1.4400    6.1200   20.1000   41.2500   31.5700
   27.4300  228.8500    9.6400  218.6100    1.2300  215.9400   17.8100   77.8500   11.0700
   52.5500   75.4800    2.0300  137.0900   52.2600    0.9000   13.1500   87.4100   17.5200
   29.1400   19.1600   12.8800    0.3000  528.2600   55.0000    6.2800   87.5800   27.1500
   26.1200  191.6500   10.7700  438.6500  122.9700  428.9900    0.8400    7.8600    2.4300
    2.1500   95.2400    0.4500   96.7400    3.9400    2.0000    0.3000   70.3200    7.3900
   27.4900   19.6400   12.8800  206.3600    2.8500   20.0200   15.0300   51.4400   20.5000
   26.5600  160.4900   10.7800    8.0500   36.0200  491.9500   20.3800   43.8700   23.2400
];

ytrain = [
     1     1     1     0     0     0     0     0     1
     0     0     0     1     1     1     1     1     0
];     

net = train(patternnet,xtrain,ytrain);

The net object can then give you predicted classifications for new observations. e.g.

ytest = net(xtest)

ytest =

      0.2543    0.1770    0.0810    0.9653    0.0458    0.3848    0.3319    0.0975    0.1332
      0.7457    0.8230    0.9190    0.0347    0.9542    0.6152    0.6681    0.9025    0.8668

If you get something that works then there are code builder tools for deployment.

PS. Try the nprtool and watch this video Pattern Recognition Tutorial

[HST20]

Thank you very much for your helpful reply. This idea i had it in my mind, but i could not able to input the data correctly to the NNTOOL, i really confused about it. Now i understood clearly about giving input to the tool. I believe that i need more input data to get more accurate results.

I have one more doubt, To implement this algorithm in a micro controller like arduino, do i need to obtain an equation from NNtool? Could you give me some hints about obtaining an equation from trained data?

[HST20]

Got some info...I need to try it. BTW, Thanks

[Greg Heath]

I suggest you try to reduce the number of inputs. The best way for classification is to use PLSREGRESS or even STEPWISEFIT(instead of PCA).

HOPE THIS HELPS.

Greg