Feature selection and transformation by machine learning reduce variable numbers and improve prediction for heart failure readmission or death

• DOI: 10.1371/journal.pone.0218760
• Publication Date: 2019-06-26T13:38:53Z
• Authors: Saqib E. Awan; Mohammed Bennamoun; Ferdous Sohel; Frank M. Sanfilippo; Benjamin J. Chow; Girish Dwivedi
• Keywords: Predictive modelling; Multilayer perceptron; Perceptron

Background The prediction of readmission or death after a hospital discharge for heart failure (HF) remains major challenge. Modern healthcare systems, electronic health records, and machine learning (ML) techniques allow us to mine data select the most significant variables (allowing reduction in number variables) without compromising performance models used death. Moreover, ML methods based on transformation may potentially further improve performance. Objective To use determine relevant also transform 30-day HF patients. Methods We identified all Western Australian patients aged 65 years above admitted between 2003–2008 linked administrative data. evaluated associated with using standard statistical selection techniques. tested new produced by original variables. developed multi-layer perceptron compared their predictive metrics such as Area Under receiver operating characteristic Curve (AUC), sensitivity specificity. Results Following discharge, proportion readmissions was 23.7% our cohort 10,757 model smaller set (n = 8) had comparable (AUC 0.62) traditional 47, AUC 0.62). Transformation 47 improved (p<0.001) 0.66). Conclusions A small selected matched that full predicting Model can be significantly transforming methods.