Machine learning-enabled high-throughput industry screening of edible oils

Long-term consumption of mixed fraudulent edible oils increases the risk of developing of chronic diseases which has been a threat to the public health globally. The complicated global supply-chain is making the industry malpractices had often gone undetected. In order to restore the confidence of consumers, traceability (and accountability) of every level in the supply chain is vital. In this work, we shown that machine learning (ML) assisted windowed spectroscopy (e.g., visible-band, infra-red band) produces high-throughput, non-destructive, and label-free authentication of edible oils (e.g., olive oils, sunflower oils), offers the feasibility for rapid analysis of large-scale industrial screening. We report achieving high-level of discriminant (AUC > 0.96) in the large-scale (n ≈ 11,500) of adulteration in olive oils. Notably, high clustering fidelity of ‘spectral fingerprints’ achieved created opportunity for (hypothesis-free) self-sustaining large database compilation which was never possible without machine learning. (137 words). •Clustering rapidly the spectral fingerprints of edible oils (in seconds) with high accuracy and high throughput manner.•This is achievable by exploiting the niche of unsupervised model which does not need a priori labeled information.•The interesting patterns reported (phenotypic variations) can be understood via the mathematics of ‘deterministic biology’