Comparative performance of artificial neural networks and support vector Machines in detecting adulteration of apple...

The detection of adulteration in apple juice concentrate is critical for ensuring product authenticity and consumer safety. This study evaluates the effectiveness of artificial neural networks (ANN) and support vector machines (SVM) in analyzing spectroscopic data to detect adulteration in apple juice concentrate. Four techniques-UV-visible, fluorescence, near-infrared (NIR) spectroscopy, and time domain ¹H nuclear magnetic resonance relaxometry (¹H NMR)-were used to generate data from both authentic and adulterated apple juice samples. Adulterants included glucose syrup, fructose syrup, grape concentrate, and date concentrate. The spectroscopic data were pre-processed and analyzed using ANN and SVM models, with performance metrics such as sensitivity, specificity, and correct classification rates (CCR) evaluated for both calibration and validation sets. Results indicated that NIR spectroscopy combined with SVM provided the highest overall accuracy, with near-perfect specificity and high CCR values, making it the most robust method for adulteration detection. UV-visible and fluorescence spectroscopy also demonstrated strong performance but were slightly less consistent across different adulterants. ¹H NMR relaxometry, while providing detailed molecular insights, showed variable sensitivity depending on the adulterant type. The findings showed the importance of selecting appropriate analytical techniques and machine learning models for food authentication. This study contributes to the development of non-destructive, rapid, and accurate methods for detecting food adulteration, which can help support industry efforts to enhance product integrity and maintain consumer trust.