I read that too much training data can compromise the performance of AI just as too less can. I'm wondering if there is a way to find out the optimal amount of training data for the algorithm?
I'm working with 3 classes and the standard temperature profile.
Hello BTSRobin. I am using 8 BME688 plates to record data of different SO2 concentrations and train an algorithm with BME AI-Studio to distinguish between high and low concentration.
The questions are:
- Is it better to use data collected over a few hours or over several days?
- Looking at the graphs of the gas data, it usually takes 1-2 hours for the signal to stabilize and change from "curved" to "straight". Do I remove that data from the beginning or use it in the training as well?