The use of data mining methods for dystocia detection in Polish Holstein-Friesian Black-and-White cattle.

OBJECTIVE: The aim of this study was to verify the usefulness of artificial neural networks (ANN), multivariate adaptive regression splines (MARS), naïve Bayes classifier (NBC), general discriminant analysis (GDA), and logistic regression (LR) for dystocia detection in Polish Holstein-Friesian Black-and-White heifers and cows and to indicate the most influential predictors of calving difficulty. METHODS: A total of 1,342 and 1,699 calving records including six categorical and four continuous predictors were used. Calving category (difficult vs easy or difficult, moderate and easy) was the dependent variable. RESULTS: The maximum sensitivity, specificity and accuracy achieved for heifers on the independent test set were 0.855 (for ANN), 0.969 (for NBC), and 0.813 (for GDA), respectively, whereas the values for cows were 0.600 (for ANN), 1.000 and 0.965 (for NBC, GDA, and LR), respectively. With the three categories of calving difficulty, the maximum overall accuracy for heifers and cows was 0.589 (for MARS) and 0.649 (for ANN), respectively. The most influential predictors for heifers were an average calving difficulty score for the dam's sire, calving age and the mean yield of the farm, where the heifer was kept, whereas for cows, these additionally included: calf sex, the difficulty of the preceding calving, and the mean daily milk yield for the preceding lactation. CONCLUSION: The potential application of the investigated models in dairy cattle farming requires, however, their further improvement in order to reduce the rate of dystocia misdiagnosis and to increase detection reliability.

Comparison between data mining methods to assess calving difficulty in cattle / Comparación entre métodos de minería de datos para evaluar la dificultad al parto en ganado / Comparação entre métodos de mineração de dados para avaliar a dificuldade no parto em bovinos

Abstract Background: Dystocia in cattle results in adverse consequences (increased calf morbidity and mortality, decreased fertility, and milk production, lower cow survival and reduced welfare) leading to considerable economic losses. Objective: To classify calvings in dairy cattle according to their difficulty using selected data mining methods (classification and regression trees (CART), chi-square automatic interaction detection trees (CHAID) and quick, unbiased, efficient, statistical trees (QUEST)), and to identify the most significant factors affecting calving difficulty. The results of data mining methods were compared with those of a more traditional generalized linear model (GLM). Methods: A total of 1,342 calving records of Polish Holstein- Friesian black-and-white heifers from four farms were used. Calving difficulty was divided into three categories (easy, moderate and difficult). Results: The percentages of calvings correctly classified by CART, CHAID, QUEST, and GLM were as follows: 35.14, 18.92, 19.82, and 43.24% (easy), 68.70, 73.91, 81.74, and 41.74% (moderate), and 77.27, 85.45, 73.64, and 81.82% (difficult), respectively. The most important factors affecting calving difficulty were bull's rank (based on the mean calving difficulty score of its daughters), calving age, farm category (based on its mean milk yield) and calving season. Conclusion: All classification models were satisfactory and could predict the class of calving difficulty.

Resumen Antecedentes: La distocia en el ganado resulta en consecuencias adversas (elevadas morbilidad y mortalidad de terneros, reducida fertilidad y producción de leche, menor supervivencia y bienestar de las vacas) que conllevan a pérdidas económicas considerables. Objetivo: Clasificar los partos del ganado lechero en función de su grado de dificultad a través de métodos seleccionados de minería de datos (árboles de clasificación y de regresión (CART), detección automática de interacción chi-cuadrado (CHAID) y árboles estadísticos no sesgados y eficientes (QUEST)) e identificar los factores más característicos de dificultad al parto. Los resultados de los métodos de minería de datos se compararon con los del modelo lineal generalizado tradicional (GLM). Métodos: Se utilizaron 1.342 registros de parto de novillas de raza polaca Holstein-Friesian blanca y negra de cuatro explotaciones lecheras. La dificultad de parto del ganado se dividió en tres categorías (fácil, moderado y difícil). Resultados: El porcentaje de partos correctamente clasificados por CART, CHAID, QUEST y GLM fue 35,14, 18,92, 19,82 y 43,24% (fácil), 68,70, 73,91, 81,74 y 41,74% (moderado), y 77,27, 85,45, 73,64 y 81,82% (difícil), respectivamente. Los factores más importantes de dificultad de parto fueron el rango de toro (determinado sobre la base de dificultad media de los partos de sus hijas), la edad al parto, la categoría de las fincas (sobre la base del rendimiento medio de leche) y la temporada de parto. Conclusión: Todos los modelos de clasificación se caracterizaron como satisfactorios y podrían predecir la clase de dificultad al parto.

Resumo Antecedentes: A distócia em bovinos resulta em consequências adversas (aumento da morbidade e mortalidade dos bezerros, diminuição da fertilidade e da produção de leite, baixa sobrevivência da vaca e redução do bem-estar) levando a consideráveis perdas econômicas. Objetivo: Classificar os partos do gado leiteiro segundo o seu grau de dificuldade através dos métodos selecionados de data mining (árvores de classificação e regressão (CART), detecção automática de interação chi-quadrado (CHAID) e ârvores estatísticas eficientes e rápidas e imparciais (QUEST)) e identificar os fatores mais importantes para a dificuldade nos partos. Os resultados dos métodos de data mining foram comparados com os resultados do modelo lineal generalizado (GLM) mais convencional. Métodos: Foram utilizados 1.342 registos de partos de novilhas da raça polaca Holstein-Frísia branca e preta de quatro fazendas. A dificuldade em um parto foi dividida em três categorias (fácil, média, difícil). Resultados: A percentagem de partos corretamente classificados através de CART, CHAID, QUEST e GLM foram de 35,14, 18,92, 19,82 e 43,24% (fácil), 68,70, 73,91, 81,74 e 41,74% (média) e 77,27, 85,45, 73,64 e 81,82% (difícil), respetivamente. Os fatores mais importantes de dificuldade no parto foram a classificação do touro (determinada com base na dificuldade média nos partos de suas filhas), a idade no momento de parto, a categoria de exploração leiteira (com base no rendimento médio de leite) e a temporada de parto. Conclusão: Todos os modelos de classificação destacaram-se por sua qualidade satisfatória e foram capazes de prever a categoria de dificuldade de um parto.