Abstract
Food texture still relies mainly on sensory tests. During the past decades, texture profile analysis (TPA) has been developed using instrumental measurements called food compression tests using a texture analyzer. Although TPA has been widely applied for a variety of foods, TPA and multi components analysis for characteristic values estimated from TPA do not work well for some foods with complicated food texture. For snacks, such as potato chips and wafers, fracture pattern varies on each piece, and the distribution of estimated values from TPA is widely spread. Machine learning (ML) may have potential to resolve the complication of the food texture analysis. Conventional TPA uses only a few (or up to 10) characteristic values. ML can handle all the data points, such as all the data force values, up to thousands of values from one compression measurement without manual selection of the important characteristic values. ML was applied to discriminate the types of potato chips and/or the other snacks, and it was demonstrated that ML could work for the discrimination of the types of snacks even for foods which are difficult to analyze allowing discrimination of foods based on TPA.
