Proficiencies of different fuzzy inference systems in predicting the production performance of broiler chickens

Abstract

Animal farming is a complex biological system because the responses of animal performance are nonlinear. In addition, thermal variables and management practices interact dynamically, making it difficult to ensure animal welfare and performance. As a result, modeling production performance under different thermal conditions is a complex task that requires predictive models. This study compared fuzzy models developed with different configurations using Mamdani and Sugeno inferences applied to the prediction of feed conversion in broilers. An experiment was conducted in four stages with a total of 240 Cobb 500 broiler chicks. The broilers were housed in climate-controlled wind tunnels and subjected to different temperatures (24, 27, 30, or 33 °C) and exposure times (1, 2, 3, or 4 days) inside cages equipped with feeders and drinkers. Feed intake and weight gain were quantified after 21 days. In both inference methods, the input variables (temperature and exposure time) were represented by both triangular and Gaussian functions. The output variable (feed conversion) was represented by singleton functions in the Sugeno inference system and by triangular and Gaussian functions in the Mamdani inference system. In addition to varying the types of membership functions in the representation of the data, all defuzzification methods of each methodology were also used. A comparison of the values predicted by each model and those obtained experimentally demonstrated that both the type of membership function and the defuzzification method influenced the final result of the prediction, with the triangular functions being better suited to the Sugeno system and the Gaussian functions being better suited to the Mamdani system for all defuzzification methods.