0
$\begingroup$

0

I would like to develop a time series classification algorithm to classify use a of parachute.

My data consist of multiple recording files (around 5min at 100hz, length of the recording vary) with a timestamp column, sensors like accelerations altitude... and a colunm label with 0 : bad, +1 good.

Exemple of data :

time;x(g);y(g);z(g)
0,005;-0,048;0;1,056
0,006;0;0;1,104
0,007;-0,048;0;1,056
0,008;0;0,048;1,104
0,009;-0,048;0;1,056
0,01;-0,048;0;1,056
0,011;-0,048;0;1,008

In the litterature and in dataset database like timeseriesclassificaction.com and UCI, datasets used for deep learning algorithms are already transformed with feature extraction with a sliding windows or if there is recording like OPPORTUNITY dataset the recording are just stacked on top of each other (like here).

I dont understand how data recording files can be processed without loosing too much information (meaning passing for a 2d array of around 5 min of reccording at 100hz to a 1d array) or without loosing information about start and end of the recording when simply merging files.

My only idea for now is to merge the time series and add a variable to differentiate them in the merged file but I am not sure it is a good idea

$\endgroup$

0

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.