Journal: Health Systems

Article Title: Can structured EHR data support clinical coding? A data mining approach

doi: 10.1080/20476965.2020.1729666

Figure Lengend Snippet: Implementation parameters for feature selection methods

Article Snippet: These feature subsets were then used to develop prediction models in a stepwise forward selection process, starting by building models using only the most relevant feature and then adding one feature to the model at a time, in decreasing order of relevance. table ft1 table-wrap mode="anchored" t5 Table 3. caption a7 Feature selection method Implementation Fast Correlation Based Filter (FCBF) • Decreasing order of relevance • 10 − 4 SU threshold • FEAST Toolbox for Matlab® (Brown, Pocock, Zhao, & Luján, 2012 ) Information Gain (IG) • Decreasing order of relevance • Forward selection – first order utility (Brown, 2009 ) Relief • Decreasing feature weight • 10 nearest neighbours • Matlab® function Chi-square • Decreasing order of χ 2 (chi-square) value Symmetrical uncertainty (SU) • Decreasing order of SU value Correlation-based Feature Selection (CFS) • Decreasing order of heuristic merit M s for each feature subset S with k features ( r c f and r f f represent the average feature-class and feature-feature correlations, respectively) (Hall & Holmes, 2003 ): M S = k ⋅ r c f ̅ ( k + k ( k − 1 ) r f f ̅ • Correlation based on symmetrical uncertainty (SU) Minimal Redundancy Maximal Relevance (mRMR) • Decreasing order of relevance • FEAST Toolbox for Matlab® (Brown et al., 2012 ) Open in a separate window Implementation parameters for feature selection methods Supervised classification models were developed and tested using Matlab® using its Statistics toolbox.

Techniques: Selection