Machine learning-based prediction of cold-pressed oil storage time using NIR and electronic-nose data

• NIR spectroscopy and e-nose data were combined with machine learning models • Fatty acids, polyphenols and volatiles of cold-pressed oils were predicted • Oil type and oxidative progression were accurately classified over six months • Multi-marker models enabled prediction of cold-pressed oil storage time Cold-pressed vegetable oils are rich in polyunsaturated fatty acids (PUFAs) and bioactive minor compounds, making them nutritionally valuable but highly susceptible to oxidative degradation. Current shelf-life estimation relies on destructive laboratory analyses and single-parameter indices, which insufficiently reflect the multi-factorial nature of lipid oxidation. This study presents a non-destructive, machine-learning (ML)-based framework to predict chemical deterioration and storage time of six cold-pressed oils (black cumin, sunflower, high-oleic sunflower, canola, linseed, and hempseed) stored for 168 days under household-relevant conditions. Target datasets comprised five fatty acids (GC-FID), 48 polyphenols (LC-MS/MS), and 18 secondary lipid oxidation products (SPME-GC-MS). Near-infrared (NIR) spectra and electronic-nose (e-nose) signals served as inputs for artificial neural network (ANN) classification and regression models. Using Bayesian regularization and Levenberg–Marquardt algorithms, fatty acids (R = 0.95), polyphenols (R = 0.97), and volatile oxidation markers (R = 0.82) were accurately predicted. Predicted multi-marker fingerprints were subsequently integrated into storage-time models using Partial Least Squares (PLS), Principal Component Regression (PCR), and Random Forest (RF). PLS captured linear deterioration trends (R = 0.87), while RF achieved the lowest prediction error (MAPE = 23%). Unlike previous NIR- or e-nose-based approaches focusing on single quality parameters, this study introduces a multi-marker framework enabling non-destructive estimation of oil storage time, supporting real-time quality monitoring and reduced food waste.