MATLAB: Organising data for machine learning using buffer function

Question

Hi,

I have some accelerometer data for various activities (standing, sitting, walking, walking upstairs, walking downstairs, laying), each activity coded by a number e.g. standing is 5 (see attached mat file – actid is the activity label and totalacc the accelerometer data). I’m looking to use the acc data to train a machine learning model to automatically identify the various postures/activities from accelerometer data.

To do so, I need to reorganise my accelerometer data into shorter buffers (50 samples long) of fixed length, for each posture/activity label. I have tried to use the buffer function but because the activities are all different sizes, I get zeros at the end (see "output_standing" variable in attached file as an example).

Is there a way to interpolate the data to replace my zeros with actual values? I tried the interp1 function but get NaN values – I think this is because it's the end of the signal, and ends in zeros.

Any help would be most appreciated!

Thanks!

Answer 1

There may be more efficient ways to create specific features for classification. See: Introduction to Feature Selection for a number of examples.

In any event, MATLAB has a number of feature selection algorithms that can make this easier and more reliable. (I have very limited experience with these functions, since they did not exist when I was doing classification, and I only looked through them out of curiosity.)