Optimizing model-agnostic random subspace ensembles

• Published in: Machine learning Vol. 113; no. 2; pp. 993 - 1042
• Main Authors: Huynh-Thu, Vân Anh, Geurts, Pierre
• Format: Journal Article Web Resource
• Language: English
• Published: New York Springer US 01-02-2024; Springer Nature B.V; Springer
• Subjects: Algorithms; Artificial Intelligence; Computer Science; Control; Engineering, computing & technology; Ensemble; Feature importances; Feature selection; Importance sampling; Ingénierie, informatique & technologie; Machine Learning; Mathematical models; Mechatronics; model-agnostic; Natural Language Processing (NLP); Optimization; Parameters; Random subspaces; Randomization; Regularization; Robotics; Sciences informatiques; Simulation and Modeling; Special Issue of the ECML PKDD 2023 Journal Track; Subspaces; Supervised learning; Feature importances; model-agnostic; Supervised learning; Ensemble
• ISSN: 0885-6125; 1573-0565; 1573-0565
• DOI: 10.1007/s10994-023-06427-5

This paper presents a model-agnostic ensemble approach for supervised learning. The proposed approach is based on a parametric version of Random Subspace, in which each base model is learned from a feature subset sampled according to a Bernoulli distribution. Parameter optimization is performed using gradient descent and is rendered tractable by using an importance sampling approach that circumvents frequent re-training of the base models after each gradient descent step. The degree of randomization in our parametric Random Subspace is thus automatically tuned through the optimization of the feature selection probabilities. This is an advantage over the standard Random Subspace approach, where the degree of randomization is controlled by a hyper-parameter. Furthermore, the optimized feature selection probabilities can be interpreted as feature importance scores. Our algorithm can also easily incorporate any differentiable regularization term to impose constraints on these importance scores. We show the good performance of the proposed approach, both in terms of prediction and feature ranking, on simulated and real-world datasets. We also show that PRS can be successfully used for the reconstruction of gene regulatory networks.