Boruta

Feature selection is a process of filtering variables with some method or criteria (Wiki). It often improves a machine learning model performance and helps with data exploration. Boruta [1] is a feature selection method that identifies all-relevant variables, instead of just selecting a minimal subset. Boruta.js is almost line-by-line port of R's package Boruta to JavaScript. It depends on the random-forest package, but can be used with other models as well.

Example

// Load boruta
const boruta = require('boruta')
 
// Generate synthetic data
const make = require('mkdata')
const [X, y] = make.friedman1({ nSamples: 1000 })
 
// Run boruta
const bor = boruta(X, y)
 
// Print results
console.log(bor.finalDecision)

Results:

{
  '0': 'Confirmed',
  '1': 'Confirmed',
  '2': 'Rejected',
  '3': 'Confirmed',
  '4': 'Rejected',
  '5': 'Rejected',
  '6': 'Rejected',
  '7': 'Rejected',
  '8': 'Rejected',
  '9': 'Rejected'
}

Web demo

You can try Boruta in the StatSim app: https://statsim.com/select/. It visualizes importance scores with final decisions and also suports multiple base models (linear regression, logistic regression, KNN, random forest)

References

1. Feature Selection with the Boruta Package (2010) Miron B. Kursa, Witold R. Rudnicki