Modeling approach for releasing a frankfurter production batch

Abstract

PurposeThe purpose of this paper is to model the relationship between 11 frankfurter physical properties and their sensory scores to classify a release of frankfurter production batches to the market.Design/methodology/approachData from 209 frankfurter batches were collected. Market batch release classifications were based on 11 physical properties via predictive and direct classification models. The predictive models under study included a regression, backpropagation neural network (BPN) and radial basis function neural network (RBFN) whereas the direct classification models were logistic regression, BPN and RBFN. Model performance was evaluated via correct classification rate.FindingsThe 11-7-4 RBFN predictive model proved superior with a 90 percent correct classification rate and 0 percent producer risk while the 11-5-1 RBFN, as a classification model, outperformed with the same level of accuracy, 90 and 0 percent, respectively. Producers prefer the less time-consuming direct classifiers for evaluation. Furthermore, the 11-5-1 RBFN direct classifier revealed that color measurement greatly influenced frankfurter batch release. Increases in redness, yellowness and brownness increased batch release probability.Originality/valueThis research attempts to establish a novel production batch release model for sausage manufacturing. Key factors can then be optimized for improving batch release probability for implementation throughout the sausage industry.