Classification and prediction dataset issues

Hi Everyone! It's the first time that I'm using the machine learning toolbox, even if I am a tireless MatLab user. (I can say that is the first time that I am facing a classification problem). I am trying to train a classifier able to " classify" the transition points relative to a certain movement. The data are from 8 inertial sensors, sampling frequency at 50 Hz. I collected the data from 17 subjects, each subject repeated the exercise ten times. I segmented the signal in epoch of 0.2 s, as seen in literature. For each epoch of each signal I computed the pursued features, saved in a matrix. This matrix is NXM, where N is the number of epoch the signal(the observations) and the M the number of features. Finally, I put together all the matrix of all exercises of all subjects. The dimensions of the final Matrix is (N*Sub*Exer)XM, where

• N is the number of epochs
• Sub is the number of subjects
• Exer is the number of exercises
• M is the number of features.

I joined this matrix with a cell array containing the labels in a table and this table was the input of my classifier. Because of it's the first time that I am tying to solve a classification problem, I used the Classification learner app. I trained my classifier(that strangely has an accuracy of 100%, so I think that I did something wrong) and I exported the model. Now, I would like to use/try this trained classifier for make predictions about a new dataset relative to an exercise of a subject. When I try I have this error:

• PTBIGMAT is the big matrix that I used for the training...

So, some questions :

• the problem is the dimension of the dataset? I should use a dataset big as the training one?
• I have to put consecutevly the epochs relative to a certain label?My current dataset:

I hope to have explained in a good way the problem. Thanks in advance for the help or also only for the attention!

Cheers,

Anna

[Don Mathis]

Hi Anna, I can't read your screenshots. They're too small on my screen. Are you following the instructions on this page:

http://www.mathworks.com/help/stats/export-classification-model-for-use-with-new-data.html ?

Make sure the table you pass to predictFcn has exactly the same variable names and variable types as the table you trained on.