Building Decision Forest via Deep Reinforcement Learning

Ensemble learning methods whose base classifier is a decision tree usually belong to the bagging or boosting. It is widely used in all aspects of machine learning and has made great achievements in classification problems. However, no previous work has ever built the ensemble classifier by maximizing long-term returns to the best of our knowledge. This paper proposes a decision forest building method called MA-H-SAC-DF (Multi-agent Hybrid Soft Actor Critic based Decision Forest) for binary classification via deep reinforcement learning. First, the building process is modeled as a decentralized partial observable Markov decision process, and a set of cooperative agents jointly constructs all base classifiers. Second, the global state and local observations are defined based on information of the parent node and the current location. Last, the state-of-the-art deep reinforcement method Hybrid SAC (Hybrid Soft Actor Critic) with hybrid action space is extended to a multi-agent system under the CTDE (centralized training decentralized execution) architecture to find an optimal decision forest building policy. The experiments indicate that MA-H-SAC-DF has the same performance as random forest, Adaboost, and GBDT (Gradient Boosting Decision Tree) on balanced datasets and outperforms state-of-the-art ensemble learning algorithms on imbalanced datasets.