Intelligent assessment of the histamine level in mackerel (Scomber australasicus) using near-infrared spectroscopy coupled with a hybrid variable selection strategy

• Published in: Food science & technology Vol. 145; p. 111524
• Main Authors: Pauline, Ong, Chang, Hsin-Tze, Tsai, I-Lin, Lin, Che-Hsuan, Chen, Suming, Chuang, Yung-Kun
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 01.06.2021
• Subjects: algorithms; Artificial neural networks; Backward interval partial least squares; data collection; Flower pollination algorithm; freshness; histamine; mackerel; near-infrared spectroscopy; Partial least squares; poisoning; prediction; risk; Scomber australasicus; Selectivity ratio; wavelet; Backward interval partial least squares; Partial least squares; Selectivity ratio; Artificial neural networks; Flower pollination algorithm
• DOI: 10.1016/j.lwt.2021.111524
• ISSN: 0023-6438

Determination of the histamine level in fish is essential not only because it is an indicator of fish freshness but also because this prevents the risk of histamine intoxication in consumers. This study used the strategy of near-infrared (NIR) spectroscopy coupled with a hybrid variable selection for rapid and nondestructive assessment of the histamine level in mackerel. To effectively identify the highly informative spectral variables, a three-step hybrid strategy, combining backward interval partial least squares, selectivity ratio and flower pollination algorithm, was developed. The optimized variables were fitted to the multivariate calibration models of partial least squares model (PLS), radial basis function neural network (RBFNN), and wavelet neural network (WNN). The best model was obtained by the optimized WNN model using the hybrid variable selection method, with R-squared (RP2) value and root mean squared error for prediction were, 0.79 and 70 mg/kg for flesh side dataset, and 0.76 and 75 mg/kg for skin side dataset. The obtained results for the skin side dataset significantly outperformed the PLS(RP2=0.58) and RBFNN (RP2=0.47) calibration models. •The proposed hybrid variable selection improved the accuracy of regression model.•The proposed hybrid variable selection reduced the complexity of regression model.•Coupling of the hybrid strategy and wavelet neural network outperformed others.•Wavelet neural network achieved R2 of 0.79 and 0.76 for flesh and skin dataset.•Near-infrared successfully predicted the histamine content in blue mackerel.