Question/Issue:
Apologies for the naive question—I’m a complete beginner in machine learning, and this is my first time using Edge Impulse.
To summarize, I’m currently getting an accuracy result of only 13.3%, and I’m hoping someone could provide guidance on how to improve it.
Here’s my setup:
- I’m using data collected with an Arduino Uno and four capacitive sensors to detect different gestures based on frequency response patterns across the four sensors (Swept Frequency Capacitive Sensing).
- The data is collected manually, reformatted in Excel, and each reading contains 640 values (160 frequency responses for each sensors, and 4 sensors are placed in different places).
- I have 10 samples per gesture for a total of 9 gestures.
From the feature explorer, it seems like the data clusters have distinct separations, but the accuracy remains poor. I’m currently using the default settings (raw data, classification).
I’d greatly appreciate any advice or insights to help me improve my results. Thank you!
Project ID:
580399