False predicts in model true features

Hello team i have this problem after training i get good result also on test data too then before live classification i give date which will suggest grade 13 but i get 1 sometimes it works fine sometimes not , here is the photo that i did 13 but got 1 feature data is in the right place.
Thank You Norik!


Would you elaborate more on your use case? what are you trying to detect? what is the dataset? and what are 1 and 13? tbh I am not able to understand what is happening in here

1 and 13 are two different data classes from my sensors, I extract features and see them with different characteristics, so it should be a good result on test data, but with live classification, I make movements similar to data class 13, but I get 1 Class

This depends on thousands of factors, how many datasets elements you have on 1 and on 13. What is the confusion matrix for both of these, and also what is the accuracy for each one of these? You can not just expect things to work because they look nice on the data explorer. Not to mention being sure if your dataset is labelled correctly.

Please share your project ID and I will have a look if you want

1 Like

yes this is
Project ID 69624


would you mind increasing the window size to 2 seconds and doing the training?

Also how many seconds do you need to do the gesture?

1 Like

yes now I will try to have gestures that are 2 seconds and there are 1 second time u 1-2 seconds

when it was only 7th grade everything worked fine, and then this started

I made the last testing file 8 and got 2, but you show everything to the confusion matrix normally and there was no overfit either

testing.3qnkur27 this is file

So to understand correctly you are doing the number gesture with your hand? and then you are collecting datasets and labelling them based on the number?

Also, are you using the same device for data collection and testing? also what kind of device are you using I see 3 accelerometers in there and I have no idea if these are necessary

I have my own device to make it simple I created classes and numbered them for example 8 this means one thing 1 another it’s not numbers it’s just a marker to make it convenient for me

my device has 5 sensors and I collect data with the same device, then I test the sensors alone and the code too

I have no idea to be honest, without knowing what are you trying to classify it would be hard for me to give you an answer, also I do not know why you have added 5 sensors. Are you sure you need them? what about 1 or two sensors? and then add a gyroscope instead of an accelerometer?

1 Like

Yes, I’m sure that everything is needed, since there are 5 fingers on the hand in this test version, I think in the near future I will transfer to the paid version, since there will be even more data, thank you