An Approach to Improving Trained Decision Tree Based Models Without Data

Abstract

The Dataless Decision Tree Improvement (D2TI) algorithm introduces a method for improving the performance of trained decision tree-based regression models without requiring access to any data. This novel approach leverages information contained within the trained decision tree to identify the decision regions. Data points are sampled from these regions in order to train a neural network that is able to better approximate the function modeled by the original decision tree. This method is demonstrated on several benchmark data sets representing varied characteristics and problem domains. A typical r2 increase of 11.4 +/- 1.5% is observed over the underlying decision tree with one outlier realizing even greater improvement. Considerations and applicability of the method are explored.

Open PDF

Document Details

Document Type
Technical Report
Publication Date
Feb 01, 2024
Accession Number
AD1221681

Entities

People

  • Benjamin Michlin
  • Jamal Rorie
  • Joshua Duclos

Organizations

  • Naval Information Warfare Center Pacific

Tags

Fields of Study

  • Computer science

Readers

  • Computational Linguistics
  • Computational Modeling and Simulation
  • Computer Networking

Technology Areas

  • AI & ML
  • AI & ML - Bayesian Inference
  • AI & ML - Machine Learning Algorithms
  • AI & ML - Neural Networks