The MultispeQ needle CO2 sensors were vital in correlating aflatoxin (AF) data to grain. However, this was a difficult effect to observe immediately. Raw readings did not seem to match perfectly with high/low AF ppb. Unfortunately, the CO2 graphs varied widely in their initial slopes because of that, I used the last 10 measurements when all samples were stabilized.
Only with ensemble methods do the CO2 readings become clear, which is not desirable for the end goal of an easy AF output method for whole grain. Analysis will continue on these samples to try and find a non-ensemble method for CO2/AF output.
The MultispeQ needle CO2, relative humidity, and temperature sensors were the best parameters for predicting fumonisin ppb from the ground grain. Adding in moisture content improved the model. The syringe method works much better than reflectance. Unsure without more ground aflatoxic samples whether respiration is better for all ground samples or just for fumonisin.
K nearest neighbors was the best algorithm almost perfectly predicting fumonisin for both the training and test data sets.