ABSTRACT
Frying is one of the oldest and simplest cooking processes, yet it comprises very complex reactions due to interaction of air, oil, and food with heat transfer operations. During deep-frying process, many chemical reactions occur, like hydrolysis, cyclization, and polymerization. These reactions release volatile and non-volatile chemicals causing undesirable changes in the oil which sometimes are harmful for human health. Thus, it is critical to analyze the oil quality from time to time during frying process. Physical parameters like color, dielectric constant, viscosity, smoke point, near-infrared (NIR) transmission spectroscopy, E-nose based, and ultrasonic technology are used for quick determination of oil quality, while free fatty acids, iodine value, saponification value, anisidine value, total polar compounds (TPM), and polymerized triacylglycerol (PTG) are more reliable chemical parameters for determining oil quality. Total phenolic content and PTG are employed as standard metrics; nonetheless, their evaluation takes a significant amount of time, effort, chemicals, and cost. Instead, these parameters are used in development of methods for NIR, E-nose, and ultrasonic technology. These methods of determining oil quality are quick, economical, reliable, easy, and non-hazardous. The majority of these approaches involved first classifying measured data, then feature extraction, and the development of regression models using machine learning (ML) methodology. Principal component analysis and hierarchical cluster analysis are commonly used to extract features, whereas support vector, partial least square, artificial neural network, and k-nearest neighbor algorithms are used to create regression models. These physical approaches combined with ML approach have potential to develop an online oil quality measurement device.
