Use este identificador para citar ou linkar para este item: http://repositorio.ufla.br/jspui/handle/1/48346
Título: Use of machine learning to predict fruit quality: a sensory study using affective scales
Título(s) alternativo(s): Uso de aprendizado de máquina na predição da qualidade de frutas: um estudo sensorial utilizando escalas afetivas
Autores: Pinheiro, Ana Carla Marques
Gularte, Márcia Arocha
Esmerino, Erick Almeida
Ferreira, Danton Diego
Vilas Boas, Eduardo Valério de Barros
Rocha, Roney Alves da
Palavras-chave: Frutas - Qualidade
Frutas - Análise sensorial
Frutas - Qualidade sensorial
Aprendizado de máquina
Escalas afetivas
Fruit - Quality
Fruits - Sensory analysis
Fruits - Sensory quality
Machine learning
Affective scales
Data do documento: 7-Out-2021
Editor: Universidade Federal de Lavras
Citação: RIBEIRO, M. N. Use of machine learning to predict fruit quality: a sensory study using affective scales. 2021. 128 p. Tese (Doutorado em Ciência dos Alimentos) – Universidade Federal de Lavras, Lavras, 2021.
Resumo: Fruit quality is one of the most important factors to ensure consistent fruit commercialization and consumer satisfaction. The role of sensory analysis in measuring consumer responses about food quality is indisputable. However, the sensory quality of fruit is generally not monitored due to time and money constraints, requiring the participation of many consumers, making sensory testing unfeasible. Thus, the present work aimed to: (i) generate mathematical models based on machine learning to predict acceptance, expectation, sweetness ideal, acidity ideal, and succulence ideal based on physicochemical data of different fruits and classify fruits according to consumer satisfaction and intention to pay more; and (ii) use different affective scales to assess sensory responses and understand the factors that influence strawberry consumer behavior. For this, in the first topic addressed in this work, the Random Forest (RF) algorithm was used to predict the sensory responses (acceptance, expectation, sweetness ideal, succulence ideal, and acidity ideal) of strawberry consumers using physical and physicochemical measurements. In addition, RF was used to classify strawberries into “satisfied” or “not satisfied” and “would pay more” or “would not pay more” based on the sensory responses of consumers. The RF obtained excellent results for the prediction task, which indicates that it is possible to correctly estimate the sensory measurements of strawberries using physical and physicochemical data. In addition, the RF was able to correctly classify the strawberry samples into the “satisfied” and “not satisfied” classes and into the “would pay more” or “would not pay more” classes. The second aspect addressed was to evaluate the use of different algorithms to predict the sensory responses of consumers of different fruits. For this, 705 orange consumers, 624 tangerine consumers, and 477 grape consumers evaluated the fruit samples according to their acceptance, expectation, ideal sweetness, succulence, and acidity, using affective scales. The results showed that there is no single algorithm that surpasses all others in predicting the consumer's response regarding the fruits evaluated through their physical, chemical, and physical-chemical parameters. In the third topic, a total of 715 consumers evaluated thirty samples of strawberries using different affective scales to quantify their acceptance, expectation, ideal sweetness, juiciness, and acidity. In addition, satisfaction and intention to pay more or not for the fruit were measured. It was observed that the attributes of sweetness, succulence and acidity directly influenced the acceptance and expectation of consumers, which consequently can generate greater satisfaction. The correlation of different affective scales can be an alternative to better understand consumer behavior. Therefore, the results indicate that the developed models can be used in fruit quality control, supporting the establishment of quality standards that consider the consumer's response. Furthermore, the proposed methodology can be extended to control the sensory quality of other fruits.
URI: http://repositorio.ufla.br/jspui/handle/1/48346
Aparece nas coleções:Ciência dos Alimentos - Doutorado (Teses)

Arquivos associados a este item:
Arquivo Descrição TamanhoFormato 
TESE_Use of machine learning to predict fruit quality: a sensory study using affective scales.pdf2,07 MBAdobe PDFVisualizar/Abrir


Os itens no repositório estão protegidos por copyright, com todos os direitos reservados, salvo quando é indicado o contrário.