My Case

I want to tackle a deep learning classification task using various smartphone sensor data. I will use a self-built data acquisition app and basically walk around with the phone, manually adding class labels to the data when necessary.

Each data instance has a timestamp and sensor values, and these instances are recorded every 1/50 sec. The labels are assigned to the corresponding timestamp of the moment I click on a button in the app's interface while recording.

The Goal

My goal is to detect ~2 sec moments in another application along with their corresponding class.

The Problem

Now I never touched time series data before and I am wondering how I should preprocess the labels before using it for a model. For now, I was thinking about having a 2sec sequence of data instances for each label (n= 50 * 2). However, I am not sure if this is the correct way of labeling in time series, as for other data types labels are assigned to a single instance.



Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.