Question/Issue: I am trying to make a ML Model for motion detection. I am collecting data on 4 motions of a device : idle, shake, pickup, and drop. I have collected data worth 3m 36s and I am getting 100% accuracy in confusion matrix. However, my pickup and drop motion often coincide with idle or shake for a fraction of a second. How can I resolve this?
Project ID: 268952
Context/Use case:
Hello @nsheth,
If it coincide with another class for a fraction of second but your window size is big enough, there should not be any issue.
I would consider adding some more data to make the model more robust.
Best,
Louis
Thank you for your response Louis. I actually realized I did not frame my question right. I notice a pickup label when I am actually dropping a package(testing sample) and then drop label when I am in fact picking a package. Is there a way to resolve this?
Hello @nsheth,
My first suggestion would be to add much more data, then make sure you don’t have a big accuracy difference between your testing accuracy and your validation accuracy (training).
If you have a big difference, then your model has probably overfit.
See here to increase your model performances: Increasing model performance - Edge Impulse Documentation
Best,
Louis
Thank you so much Louis for all the help. I think my model is getting more accurate now I have fetched it more data and have changed a few parameters to fine tune it.
