Machines that learn – some recent history

Statistical learning machines arose as a branch of computer science. These intriguing computer-intensive methods are now being applied to extract useful knowledge from an increasingly wide variety of problems involving oceans of information, heterogeneous variables, and analytically recalcitrant data. Such problems have included:

• predicting fire severity in the US Pacific Northwest (Holden et al., 2008);

• predicting rainfall in Northeastern Thailand (Ingsrisawang et al., 2008);

• handwriting recognition (Schölkopf and Smola, 2002);

• speech emotion classification (Casale et al., 2008).

Learning machines have also been applied to biomedical problems, such as:

• colon cancer detection derived from 3-D virtual colonoscopies (Jerebko et al., 2003, 2005);

• detecting differential gene expression in microarrays, for data that can involve more than a million single nucleotide polymorphisms (SNPs), in addition to clinical information, over several thousand patients (Díaz-Uriarte and Alvarez de Andrés, 2006);

• predicting short-term hospital survival in lupus patients (Ward et al., 2006);

• finding the most predictive clinical or demographic features for patients having spinal arthritis (Ward et al., 2009).