Healthcare Data Analytics

202 Healthcare Data Analytics

(ii) smaller numb ers of markers used for the predictive model allow the design of the diagnostic

devices for cheaper and faster prediction;

(iii) pre dictive mod els with fewer markers can sugge st biological interpretation that may poten-

tially lead to better un derstanding of the molecular underpinnings of prognosis.

The problem of feature selection has been thoroughly studied in the statistics and machine learn-

ing communities [59]. In order to select a subset of the fe atures G, called signatu re, which comprise

q features selected from p candidates, the simplest way is to calculate a discriminative score for

each ...