Early classification of timeseries

[albertoazzari]

Hi,
I'm trying to figure out how to solve this problem that I'll try to explain in the next few lines.
I have a timeseries of length ~200k values and every 700 points I have a label that indicates the status of the signal.
In this medical problem every 700 points I know if the signal is in a pathological or healthy status. The goal of the framework I want to implement is to understand if I can correctly predict, early in time, if the signal is going to be pathological in order to raise warning.

I have tried different approach so far, but none of them really works. For example I have tried to classify using a sliding window of size 700 and predict the label in the "classic" way with different classifier. In order to implement early prediction I simply shifted the labels backwards (i.e. 1st sliding window in input and label of the next 700 point). Actually I did not consider the signals as a timeseries but always as vectors not related in time.

Can someone give me some tips on how i can solve this problem considering the time factor ?