Feature Selection for Regression Based on Gamma Test Nested Monte Carlo Tree Search.

Area of Science:

• Machine Learning
• Data Mining
• Statistical Modeling

Background:

• Feature selection is crucial for improving regression model performance and interpretability.
• Existing methods may struggle with complex datasets or computational efficiency.

Purpose of the Study:

• To investigate the efficacy of Nested Monte Carlo Tree Search (NMCTS) for feature selection in regression.
• To introduce and evaluate a novel GNMCTS method incorporating the Gamma test for enhanced performance.

Main Methods:

• Nested Monte Carlo Tree Search (NMCTS) framework adapted for feature selection.
• Integration of the Gamma test as a reward function within the NMCTS simulation.
• Combination of Gamma test's Vratio with UCT-tuned and modified stopping conditions.

Main Results:

• The proposed GNMCTS method demonstrated superior performance compared to vanilla MCTS and other feature selection techniques.
• GNMCTS effectively preserves relevant information within the original feature space.
• The method proved robust and efficient across seven diverse numeric datasets.

Conclusions:

• GNMCTS is a robust and effective algorithm for feature selection in regression tasks.
• The approach achieves good performance within a reasonable computational budget.
• This method offers a promising alternative for complex feature selection problems.