Aplicação De Redes Neurais Artificiais Como Teste De Detecção De Fraude De Leite Por Adição De Soro De Queijo

Abstract

This study aimed to employ Artificial Neural Networks to classify milk samples from routine analysis of a dairy company in order to identify adulteration by whey addition. The samples were prepared by mixing the milk with different whey concentrations (0, 1, 5, 10, and 20%), which were then analyzed for temperature, fat content, solids-non-fat, bulk density, protein, lactose, minerals, freezing point, conductivity, and pH, for a total of 167 assays. Out of these, 101 were used to train the network, 33 for validation, and 33 to test the artificial neural network. The best classification was obtained using a radial basis function neural network. k-means algorithm was used to obtain the network center, k-nearest was used to define the receptive fields, and the pseudo-inverse method was used to define the weights of the output layer. The best result was found with a network with 10 neurons in the input layer, 40 neurons in the hidden layer, and two neurons in the output layer, achieving over 95% accuracy in classification. The classification methodology using artificial neural networks has strong potential to be applied in interpreting data from routine analysis in dairy companies in order to classify milk adulterated with whey and, later, confirm the result using official methodologies.

Artificial Neural Networks for Predicting Animal Thermal Comfort

“…The THI was estimated using the equation proposed by Thom (1959) The THI values on each day of the year in the Julian calendar were input and processed using the statistical program R to define the possible architectures of artificial neural networks -ANNs (R CORE TEAM, 2018). Taking as a reference the studies by Binoti et al (2014a, b), Valente et al (2014), and Borges et al (2017), multilayer perceptron ANNs were chosen for predicting this index as a function of the day of the year. After that, several network architectures were defined using the Julian day as the independent input variable and the THI as the dependent output variable, in addition to one or two intermediate layers with up to 25 neurons each.…”

“…Cansian et al 2014, Georgens et al (2014), Valente et al (2014), and Borges et al (2017) reported that the dataset was randomly divided into two subsets for network training and data validation, corresponding to 75% and 25%, respectively. In all architectures, the logistic function was used to activate the networks, and initial weights between -0.5 and +0.5 were randomly generated.…”

Artificial Neural Networks Applied in Forest Biometrics and Modeling: State of the Art (January/2007 to July/2018)

“…This traditional computational tool is capable of obtaining excellent results in different areas and applications, as described in the work of Oliveira et al (2015) and Steiner et al (2006), regarding bank lending through clients' classification (defaulters or paying customers). Other works in this respect are those of Valente et al (2014), concerning solving problems involving adulteration in the milk manufacturing process, and Gonçalves et al (2016) on the classification of forest strata, based on remote sense data. These are just some of the practical examples that could be mentioned.…”

Redes neurales artificiales para estimar las pérdidas en la producción lechera

“…Basado en los trabajos de Terra & Passador (2012), Ventura et al (2012), Georgens et al (2014) y Valente et al (2014), se optó por las redes neurales artificiales con la estructura "Perceptrons" de múltiples camadas (MPL). Para definir la arquitectura se varió el número de camadas intermediarias y el número de neuronas en las camadas, conforme los criterios de Terra & Passador (2012) y Ventura et al (2012).…”