Skip to main content

Table 2 Most prominent feature subset selected using multivariate features analysis based on Weka correlation-based feature selection method with best fit search method for classification of not frail and pre-frail (78 features)

From: Machine learning models for identifying pre-frailty in community dwelling older adults

Fried Frailty Phenotype

Clinical Frailty Scale

Features

Scorea

Features

Score

Lower 6 Minute Walk Test (6MWT) score (metres walked)

0.528

Higher Body Mass Index (BMI)

0.179

Weaker grip strength (right hand, sitting)

0.366

Lower 6MWT score (metres walked)

0.162

Increased likelihood of unintended weight loss of more than 5 kg

0.058

Fewer hours of vigorous activity (past week)

0.127

Higher Kessler Psychological Distress Scale (K10) score

0.025

Higher K10 score (distress)

0.091

Housing type (more likely to live in a house)

0.023

Lower functional mobility score (hurdle, left leg first)

0.088

 

Higher K9 score (distress)

0.069

Shortness of breath

0.051

Diagnosed with any health conditions

0.050

Lower functional mobility score (raised right leg/left arm)

0.047

Fewer minutes gardening in the past week

0.038

Less able to walk up and down 15 stairs without rest

0.032

Fewer minutes of vigorous activity (past week)

0.031

Poorer balance assessment (eyes open)

0.021

Poorer balance assessment (heel toe steps backwards)

0.016

  1. Bold text indicates features found in both the FFP and CFS feature ranking
  2. aScore: variable ranking by their contribution to the prediction using Random Forest algorithm. Higher scores indicate greater predictive value