Monte Carlo Tree Search-Based Recursive Algorithm for Feature Selection in High-Dimensional Datasets.

Area of Science:

• Data Science
• Machine Learning
• Artificial Intelligence

Background:

• Big data presents challenges due to complexity and high dimensionality.
• Feature selection is crucial for dimensionality reduction and improving model performance.
• Existing Monte Carlo Tree Search (MCTS) methods for feature selection face a tradeoff between search depth and simulation count.

Purpose of the Study:

• To propose a new iterative algorithm for feature selection that overcomes limitations of existing MCTS-based methods.
• To improve the balance between classification accuracy and dimensionality reduction.
• To enhance the efficiency of feature selection in high-dimensional big data.

Main Methods:

• A novel algorithm is proposed that iteratively builds multiple feature selection trees in a recursive manner.
• Each successor tree has a reduced state space, intensifying the impact of tree search.
• The number of Monte Carlo Tree Search (MCTS) simulations is kept fixed while improving search effectiveness.

Main Results:

• Experiments were conducted on 16 benchmark datasets to validate the proposed algorithm.
• The algorithm was compared against state-of-the-art methods in terms of classification accuracy and feature selection ratio.
• The iterative approach demonstrated improved performance in selecting optimal feature subsets.

Conclusions:

• The proposed iterative multi-tree approach effectively addresses the limitations of single-tree MCTS methods in feature selection.
• This method offers a promising solution for high-dimensional big data by optimizing the feature selection process.
• The algorithm achieves a better compromise between classification accuracy and reduced dimensionality, outperforming existing techniques.