Predictive Modelling Based on Statistical Learning in Biomedicine

Twenty years ago, the journal Computational and Mathematical Methods in Medicine was launched under its previous title Journal of Theoretical Medicine. During those years at the end of the last century, the understanding of machine learning technology and its potential combination with statistical modelling approaches was in its infancy. The modern term “statistical learning” for this fusion of methodology from different scientific areas could already be found in the scientific literature (see Vapnik [1, 2]), but its meaning was slightly different from today. The famous textbook by Hastie et al. [3] popularised the term in its current meaning when being published in its first edition in 2001. During recent years, considerable research has been devoted to exploring this combination of state-of-the-art statistical methodology with machine learning techniques. Such an approach provides many practical advantages, particularly regarding data situations frequently encountered in modern biomedical research characterized by large numbers of potential features or variables. In such situations, the primary aim is often to obtain sparse and explanatory models, which can be generalized effectively. Via statistical learning approaches, interpretable prediction rules leading to accurate forecasts for future or unseen observations can be deduced from potentially high-dimensional data.