Predictive Modeling for Large Data Sets and One Health.

Prediction

Modeling Associations between (risk) factors and health or production outcomes are often modeled using parameter estimation models and prediction models from the regression family of probabilistic models. Goodness of fit for parameter estimation models can be quantified and performance parameters for prediction models compared to optimize the models (1). Prediction modeling relates to the process of “Statistical Learning” (2,3). Current developments of prediction models have profound consequences for how humans, animals, plants and the environment function. Prediction models currently applied to large data sets and Artificial Intelligence influence definitions for health and wellbeing. It is my opinion that current One Health concepts need to include to concept of “Cyber Health” and the healthcare of “Cyber Talent”.

Statistical Learning and Machine Learning

Statistical learning processes have resulted in Machine Learning Algorithms that include many types of models among which linear models such as generalized linear models (GLM), lasso and elastic-net regularized generalized linear models (GLMNET), linear discriminant analysis (LDA), linear support vector machines (SVM), nearest neighbor methods (KNN), naive Bayes (NB), classification trees (CART), neural networks (NNET), gradient boosting machine (GBM), random forests (RF) to name a few (2,4). “Statistical Learning” can be “supervised” if an outcome is known or “unsupervised” if no outcome is available to result, for example, in clustering of data (5).

Deep Learning

In 2015, Google released TensorFlow, a deep learning platform that promotes network modeling approaches to multiple outcomes, speech, image, video and (hand-)writing computer recognition, genetic and -omics data among many other types of data sets (6,7)(Chollet 2018, Pattanayak 2017). Examples for alternative “backends” to TensorFlow are: PyTorch (see Pytorch.org), Theano(8), MXNet(9) and CNTK(10). Software platforms written in for example C++(11), Python(12) or for R(13) that handle such modeling “backends” are: Keras(14), TPOT(15), scikit-learn(16), H2O(17), Azure(18) among others. The latter four examples can handle Artificial Intelligence, computer vision and deep learning modeling approaches as part of machine learning efforts for predictions from large data sets. More freeware and licensed platforms are growing in numbers as we speak.

Large Data Sets

With the advent of increasing data volumes, more freeware, pipelines and computing capacity, a process termed “democratization of deep learning” (6) has been initiated that promotes access for everyone to state-of-the-art machine learning approaches including data mining, dimension reduction and parameter shrinkage. It is not surprising that large data sets reach across species into prediction modeling of One Health outcomes connected to environmental, animal, human and “Cyber [...]