Using multiple regression, neural networks and supprot vector machines to predict lamb carcasses composition

• Autor(a) principal: Silva, Filipe
• Data de Publicação: 2010
• Outros Autores: Cortez, Paulo, Cadavez, Vasco
• Idioma: eng
• Título da fonte: Repositórios Científicos de Acesso Aberto de Portugal (RCAAP)
• Texto Completo: http://hdl.handle.net/10198/2690
• topic: Carcass; Multiple regression; Neural networks; Support vector machines; Tissue

The objective of this work was to use a Data Mining (DM) approach to predict, using as predictors the carcass measurements taken at slaughter line, the composition of lamb carcasses. One hundred and twenty ve lambs of Churra Galega Bragan cana breed were slaughtered. During carcasses quartering, a caliper was used to perform subcutaneous fat measurements, over the maximum depth of longissimus muscle (LM), between the 12th and 13th ribs (C12), and between the 1st and 2nd lumbar vertebrae (C1). The Muscle (MP), Bone (BP), Subcutaneous Fat (SFP), Inter-Muscular Fat (IFP), and Kidney Knob and Channel Fat (KKCF) proportions of lamb carcasses were computed. We used the rminer R library and compared three regression techniques: Multiple Regression (MR), Neural Networks (NN) and Support Vector Machines (SVM). The SVM model provided the lowest relative absolute error for the prediction of BP, SFP and KKCF, while MR presented the best predictions for MP and IFP. Also, a sensitivity analysis procedure revealed the C12 measurement as the most relevant predictor for all ve carcass tissues.