I want to use unlabeled sensor data for classification.

In a copy project of a paper, I'm trying to implement a sensor in my ear to recognize facial expressions (output is 0 when I'm not doing anything, 1 when I open my mouth wide).

Easy implementation of neural networks in the chain (for beginners)

Using the above site as a reference, we studied with a csv file that summarizes the labeled sensor data (photo reflector, 0.1s*20=2s).
Now, I'm trying to get that learned model to input 2 seconds of data from Arduino in real time (although not strictly real time) via serial communication for facial expression recognition.
However, only the data labeled as the code of the above site will emit the output.
In other words, you can only answer the labeled data to determine if the labeled data is correct.

How do I modify the code and use data with only unlabeled sensor values to output which expression it is?

Even if it's not a value from Arduino, if you can read and classify unlabeled csv files, you can change the csv part to Arduino's, so I think it's good to read unlabeled data anyway.

By the way, if you try to read 20 columns of csv data per row in the current state, you will be asked for another column of data (that means you don't have enough columns of label values).