Search

The huge amount of data produced by the Internet of Things need to be validated and curated to be prepared for the selection of relevant data in order to prototype models, train them, and serve the model. On the other side, blockchains and open data are also important data sources that need to be integrated into the proposed integrative models. It is difficult to find a sufficiently versatile and agnostic architecture based on the main machine learning frameworks that facilitate model development and allow continuous training to continuously improve them from the data streams. The paper describes the conceptualization, implementation, and testing of a new architecture that proposes a use case agnostic processing chain. The proposed architecture is mainly built around the Apache Submarine, an unified Machine Learning platform that facilitates the training and deployment of algorithms. Here, Internet of Things data are collected and formatted at the edge level. They are then processed and validated at the fog level. On the other hand, open data and blockchain data via Blockchain Access Layer are directly processed at the cloud level. Finally, the data are preprocessed to feed scalable machine learning algorithms. [ABSTRACT FROM AUTHOR]

To develop a breed assignment model, three main steps are generally followed: 1) The selection of breed informative single nucleotide polymorphism (SNP); 2) The training of a model, based on a reference population, that allows to classify animals to their breed of origin; and 3) The validation of the developed model on external animals i.e., that were not used in previous steps. However, there is no consensus in the literature about which methodology to follow for the first step, nor about the number of SNP to be selected. This can raise many questions when developing the model and lead to the use of sophisticated methodologies for selecting SNP (e.g., with iterative algorithms, partitions of SNP, or combination of several methods). Therefore, it may be of interest to avoid the first step by the use of all the available SNP. For this purpose, we propose the use of a genomic relationship matrix (GRM), combined or not with a machine learning method, for breed assignment. We compared it with a previously developed model based on selected informative SNP. Four methodologies were investigated: 1) The PLS_NSC methodology: selection of SNP based on a partial least square-discriminant analysis (PLS-DA) and breed assignment by classification based on the nearest shrunken centroids (NSC) method; 2) Breed assignment based on the highest mean relatedness of an animal to the reference populations of each breed (referred to mean_GRM); 3) Breed assignment based on the highest SD of the relatedness of an animal to the reference populations of each breed (referred to SD_GRM) and 4) The GRM_SVM methodology: the use of means and SD of the relatedness defined in mean_GRM and SD_GRM methodologies combined with the linear support vector machine (SVM), a machine learning method used for classification. Regarding mean global accuracies, results showed that the use of mean_GRM or GRM_SVM was not significantly different (Bonferroni corrected P > 0.0083) than the model based on a reduced SNP panel (PLS_NSC). Moreover, the mean_GRM and GRM_SVM methodology were more efficient than PLS_NSC as it was faster to compute. Therefore, it is possible to bypass the selection of SNP and, by the use of a GRM, to develop an efficient breed assignment model. In routine, we recommend the use of GRM_SVM over mean_GRM as it gave a slightly increased global accuracy, which can help endangered breeds to be maintained. The script to execute the different methodologies can be accessed on [ABSTRACT FROM AUTHOR]

Simple Summary: First, the current work consisted of validating the feasibility of large-scale dairy cow bodyweight prediction from models involving the day in milk, milk yield, parity, and milk mid-infrared spectrum. Second, it aimed to improve the accuracy of predictive models by using feature selection algorithms to decrease the number of predictors to limit overfitting. The models, using accessible and low-cost measurements, provided highly reproducible predictions. These could be easily obtained on an individual basis throughout a cow's productive life by dairy herd improvement organizations, thus providing potentially relevant information for the dairy farmer at three levels: economics (reproductive performance), animal welfare (disease detection), and environment (methane production). Knowing the body weight (BW) of a cow at a specific moment or measuring its changes through time is of interest for management purposes. The current work aimed to validate the feasibility of predicting BW using the day in milk, parity, milk yield, and milk mid-infrared (MIR) spectrum from a multiple-country dataset and reduce the number of predictors to limit the risk of over-fitting and potentially improve its accuracy. The BW modeling procedure involved feature selections and herd-independent validation in identifying the most interesting subsets of predictors and then external validation of the models. From 1849 records collected in 9 herds from 360 Holstein cows, the best performing models achieved a root mean square error (RMSE) for the herd-independent validation between 52 ± 2.34 kg to 56 ± 3.16 kg, including from 5 to 62 predictors. Among these models, three performed remarkably well in external validation using an independent dataset (N = 4067), resulting in RMSE ranging from 52 to 56 kg. The results suggest that multiple optimal BW predictive models coexist due to the high correlations between adjacent spectral points. [ABSTRACT FROM AUTHOR]