Parameter tuning problem

I have some questions about parameter tuning that I need to ask
in the project 126809 and the project 126766, there is some gap between my training accuracy and test accuracy. It seems that there is an overfitting phenomenon. Do you have any experience in adjusting the parameters of edge impulse? Especially the second project, the overfitting is a bit serious, but no matter how I simplify the model structure, the problem is not solved in the end.

Hi @sherry,

It looks like you have a dataset that has a lot of overlap in the different features among the classes. This will make it difficult for any ML algorithm to discern the difference between classes (regardless of the hyperparameters you set for the model). Many of the different sounds you picked up simply end up looking like a version of noise with similar frequencies. You might have to rethink about the problem you are solving and the type of data you are collecting.