A comparative evaluation of the generalised predictive ability of eight machine learning algorithms across ten clinical metabolomics data sets for binary classification

The multidisciplinary field of data science is concerned with extracting insights from data using a diverse set of computational methodologies, theories, and technologies (Blei and Smyth 2017). Within data science, there are two competing scientific philosophies: classical statistics and machine learning (Breiman 2001b). Classical statistics aims to formalise relationships between dependent and independent variables based on a clearly defined set of assumptions from which mathematical models are parametrised. The aim is to derive meaningful statistical inference (properties of an underlying probability distribution) for the measured variables, assuming that the observed data is sampled from a larger population. Conversely, machine learning uses ad-hoc computational algorithms that iteratively optimise (or'learn') without necessarily relying on any formal statistical assumptions (Bishop 1995).