Multivariate Analysis for Data Mining to Characterize Broiler House Environment in Winter

Abstract

Studies of environment in confined animal production systems have generated massive, high-dimensional datasets, and thus robust and practical methods for mining these complex datasets are critical in the context of precision (smart) livestock farming (PLF). This study explored the potential of multivariate analysis tools for analyzing environmental data in a broiler house. An experiment was conducted to collect a comprehensive set of environmental data, including TSP, PM10, PM2.5, NH3, CO2, air temperature, relative humidity, and in-cage and aisle wind speed, at 60 locations in a typical commercial broiler house. The dataset was divided into three bird growth phases (weeks 1-3, 4-6, and 7-9). The Spearman correlation analysis and principal component analysis (PCA) were used to investigate latent associations between different environmental variables, based on which the variables that played important roles in broiler house environment were identified. Three methods of cluster analysis, namely K-means, K-medoids, and fuzzy c-means cluster analysis (FCM), were evaluated for grouping the measured parameters in terms of their effects on the environment in the broiler house. In general, the results of Spearman and PCA indicated that the in-cage wind speed, aisle wind speed, and relative humidity played a critical role in environment during broiler rearing. All three clustering methods were shown to be suitable for grouping data, with FCM performing slightly better than the other two methods. Based on data clustering, the broiler house space was divided into three, two, and two subspaces (clusters) for weeks 1-3, 4-6, and 7-9, respectively. The subspace in the middle of the house had poorer air quality than other subspaces.

Keywords: broiler house, microclimate, air quality, multivariate analysis, data mining