Fair and Optimal Decision Trees: A Dynamic Programming Approach

被引:0
|
作者
van der Linden, Jacobus G. M. [1 ]
de Weerdt, Mathijs M. [1 ]
Demirovic, Emir [1 ]
机构
[1] Delft Univ Technol, Dept Comp Sci, Delft, Netherlands
关键词
D O I
暂无
中图分类号
TP18 [人工智能理论];
学科分类号
081104 ; 0812 ; 0835 ; 1405 ;
摘要
Interpretable and fair machine learning models are required for many applications, such as credit assessment and in criminal justice. Decision trees offer this interpretability, especially when they are small. Optimal decision trees are of particular interest because they offer the best performance possible for a given size. However, state-of-the-art algorithms for fair and optimal decision trees have scalability issues, often requiring several hours to find such trees even for small datasets. Previous research has shown that dynamic programming (DP) performs well for optimizing decision trees because it can exploit the tree structure. However, adding a global fairness constraint to a DP approach is not straightforward, because the global constraint violates the condition that subproblems should be independent. We show how such a constraint can be incorporated by introducing upper and lower bounds on final fairness values for partial solutions of subproblems, which enables early comparison and pruning. Our results show that our model can find fair and optimal trees several orders of magnitude faster than previous methods, and now also for larger datasets that were previously beyond reach. Moreover, we show that with this substantial improvement our method can find the full Pareto front in the trade-off between accuracy and fairness.
引用
收藏
页数:13
相关论文
共 50 条
  • [1] Optimal Survival Trees: A Dynamic Programming Approach
    Huisman, Tim
    van der Linden, Jacobus G. M.
    Demirovic, Emir
    THIRTY-EIGHTH AAAI CONFERENCE ON ARTIFICIAL INTELLIGENCE, VOL 38 NO 11, 2024, : 12680 - 12688
  • [2] MurTree: Optimal Decision Trees via Dynamic Programming and Search
    Demirovic, Emir
    Lukina, Anna
    Hebrard, Emmanuel
    Chan, Jeffrey
    Bailey, James
    Leckie, Christopher
    Ramamohanarao, Kotagiri
    Stuckey, Peter J.
    JOURNAL OF MACHINE LEARNING RESEARCH, 2022, 23
  • [3] MurTree: Optimal Decision Trees via Dynamic Programming and Search
    Demirović, Emir
    Lukina, Anna
    Hebrard, Emmanuel
    Chan, Jeffrey
    Bailey, James
    Leckie, Christopher
    Ramamohanarao, Kotagiri
    Stuckey, Peter J.
    Journal of Machine Learning Research, 2022, 23
  • [4] Optimization and analysis of decision trees and rules: dynamic programming approach
    Alkhalid, Abdulaziz
    Amin, Talha
    Chikalov, Igor
    Hussain, Shahid
    Moshkov, Mikhail
    Zielosko, Beata
    INTERNATIONAL JOURNAL OF GENERAL SYSTEMS, 2013, 42 (06) : 614 - 634
  • [5] Necessary and Sufficient Conditions for Optimal Decision Trees using Dynamic Programming
    van der Linden, Jacobus G. M.
    de Weerdt, Mathijs M.
    Demirovic, Emir
    ADVANCES IN NEURAL INFORMATION PROCESSING SYSTEMS 36 (NEURIPS 2023), 2023,
  • [6] A multi-objective genetic programming approach to developing Pareto optimal decision trees
    Zhao, Huimin
    DECISION SUPPORT SYSTEMS, 2007, 43 (03) : 809 - 826
  • [7] Learning optimal decision trees using constraint programming
    Verhaeghe, Helene
    Nijssen, Siegfried
    Pesant, Gilles
    Quimper, Claude-Guy
    Schaus, Pierre
    CONSTRAINTS, 2020, 25 (3-4) : 226 - 250
  • [8] Learning Optimal Decision Trees using Constraint Programming
    Verhaeghe, Helene
    Nijssen, Siegfried
    Pesant, Gilles
    Quimper, Claude-Guy
    Schaus, Pierre
    PROCEEDINGS OF THE TWENTY-NINTH INTERNATIONAL JOINT CONFERENCE ON ARTIFICIAL INTELLIGENCE, 2020, : 4765 - 4769
  • [9] Learning optimal decision trees using constraint programming
    Hélène Verhaeghe
    Siegfried Nijssen
    Gilles Pesant
    Claude-Guy Quimper
    Pierre Schaus
    Constraints, 2020, 25 : 226 - 250
  • [10] A dynamic programming based pruning method for decision trees
    Li, XB
    Sweigart, J
    Teng, J
    Donohue, J
    Thombs, L
    INFORMS JOURNAL ON COMPUTING, 2001, 13 (04) : 332 - 344