Potential of low frequency dielectric spectroscopy and machine learning methods for extra virgin olive oils discrimination based on the olive cultivar and ripening stage

Olive cultivars present unique chemical properties due to the different ripening conditions and the dielectric factor of the extracted extra virgin olive oil (EVOO) is one of the most acceptable parameters for their evaluation. This study aims at discriminating olive oil samples using their chemical and dielectric properties. Three cultivars of olives: Oily, Yellow and Fishemi, at three ripening stages of unripe, semi-ripe and ripe, harvested using two different machines were tested. The quality characteristics of olive oil include the value of acidity, peroxide, sterol compounds, fatty acid composition and total phenol of the extracted olive oil were measured. The EVOO’s dielectric parameters were measured using a laboratory developed low-frequency device. The measures were done in the range from 0.1 to 10 MHz. The obtained parameters consisted of gain and phase shift voltages that were analyzed by principal component analysis, linear discriminant analysis, decision trees, support vector machine and artificial neural network. Out-of sample validation indicated the artificial neural network had the best performance with correlation coefficient of 0.9479, mean absolute error of 4.762 and root mean square error of 6.6652.The results revealed that the sensor measure followed by a machine learning approach have potential for the industrial application for discriminating the EVOO on the basis of the cultivar and ripening stage of the processed olive.