Electronic nose combined with chemometric approaches to assess authenticity and adulteration of sausages by soy protein

This paper is devoted to the multiclass classification of quartz crystal microbalance sensor-array data based on two chemometric approaches, and aimed at the development of rapid analytical technique (both at the chemical and data levels) for detection and quantitative assessment of soy protein at the levels 0, 10, 20, 30% w/w in sausages for avoiding different type of frauds and thus providing the effective meat products quality control issue. The first approach was based on the simple algorithm for odor pattern recognition with the use of geometric parameters (new star coefficient and perimeter), which allows assessing authenticity and adulteration of four sausage types with a high degree of reliability using identification criteria. While the second approach was based on electronic nose features as input vectors for the optimized probabilistic neural network (PNN). The area values and maximum response values were extracted as features from the electronic nose responses for evaluation and comparison their ability to discriminate the four classes of sausages, and the effectiveness of data pre-processing and compression (normalizing, autoscaling, PCA) was investigated. The reliability of classification of 100% was obtained using raw maximum response values as input vectors of PNN that allows to predict correctly all sausage types for avoiding a labelling fraud.