MCES: a novel Monte Carlo evaluative selection approach for objective feature selections.

Area of Science:

• Machine Learning
• Data Mining
• Computational Statistics

Background:

• Feature selection research has predominantly focused on classification tasks, neglecting nonlinear system estimations.
• Limited computationally efficient feature selection methods exist for nonlinear regression contexts.

Purpose of the Study:

• To propose a novel, computationally efficient feature selection approach for nonlinear system estimations.
• To develop a method applicable to both classification and nonlinear regression tasks.

Main Methods:

• Introduced the Monte Carlo evaluative selection (MCES), an objective sampling method for relevancy measure estimation.
• Designed MCES to be independent of specific underlying induction algorithms.

Main Results:

• MCES demonstrated effectiveness across two classification and two regression tasks.
• The method successfully identified correlated and irrelevant features using weight ranking.
• MCES proved applicable to both nonlinear system estimation and classification.

Conclusions:

• MCES offers a versatile and efficient feature selection solution for both classification and nonlinear regression.
• The algorithm's independence from induction algorithms enhances its broad applicability.
• MCES provides a robust method for improving model accuracy by selecting relevant features.