Rapid detection of total phenolics, antioxidant activity and ascorbic acid of dried apples by chemometric algorithms

Abstract

Multivariate approaches like machine learning are commonly used in estimation of biochemical traits from spectral and color characteristics of foodstuffs and agricultural commodities. In present study, windfall apples of Golden Delicious, Oregon Spur and Granny Smith cultivars were dried in open-sun, controlled greenhouse, microwave oven (200W), hybrid system (100W + 60°C), convective dryer (70°C) and freeze-dryer (−55°C). Spectral, chromatic and biochemical characteristics of dried apples were determined and assessed through machine learning algorithms. Total phenolic matter, DPPH (2,2-Diphenyl-1-picrylhydrazyl), FRAP (Ferric Reducing Antioxidant Power) and ascorbic acid content were estimated with the use of five different machine learning algorithms (artificial neural networks, k-nearest neighbor, random forest, gaussian processes and support vector regression). The most successful results were achieved in estimation of total phenolic content (R ≥ 0.85). Additionally, Multilayer Perceptron, Support Vector Regression and Gaussian Processes were identified as the best machine learning algorithms in estimation of biochemical compositions of dried apples.