Application of Support Vector Classification Algorithms for the Prediction of Quality Level of Frozen Shrimps (Litopenaeus vannamei) Suitable for Sensor-Based Time-Temperature Monitoring

作者：Ahmad Imran^*; Jeenanunta Chawalit

来源：Food and Bioprocess Technology, 2015, 8(1): 134-147.

DOI：10.1007/s11947-014-1377-8

• Frozen shrimp
• Quality classification
• SVC
• SVM
• Quality management
• Shelf-life
• Cold chain

Classification of frozen shrimps (Litopenaeus vannamei) at the end-of-storage period in terms of physical, chemical, microbial, and sensory attributes is demonstrated in this paper. The quality parameters of frozen shrimps stored in isothermal and fluctuating temperature conditions were found significantly different (p < 0.05). In order to classify products into binary classes of quality due to abusive and non-abusive handling (1, -1), useful features of the time-temperature curve were extracted to train the network using support vector classification (SVC). It was generally possible to classify shrimps based on end-of-storage quality by the available kernels; however, the radial basis RBF) was found to be the most suitable in terms of best fit and lower cost of calibration. The evaluation of the SVC kernels was done in terms of sensitivity and specificity values which were 0.962 and 0.992 in the case of RBF, respectively, whereas other kernels, polynomial and sigmoid, showed signs of over-fitting. The classification models were selected on the basis of root mean squared error (RMSE), mean absolute error (MAE), and other standard error measuring procedures. An independent cross validation procedure of the developed models was performed using a different set of data (n = 18) against the model predicted values which lower root mean squared error (RMSE) terms (0.026-2.44) for all the quality indices, which confirmed the validity of the SVC technique.

• 出版日期2015-1