Rough sets and Laplacian score based cost-sensitive feature selection.

Area of Science:

• Machine Learning
• Data Mining
• Artificial Intelligence

Background:

• Cost-sensitive feature selection is crucial for preprocessing in machine learning.
• Existing heuristic algorithms often overlook inter-feature relationships, impacting efficiency.

Purpose of the Study:

• To propose a novel cost-sensitive feature selection algorithm.
• To address limitations of heuristic methods by incorporating feature relationships.

Main Methods:

• Developed a new algorithm combining rough sets and Laplacian score for feature importance evaluation.
• Evaluated feature importance considering both individual feature characteristics and their local relationships.
• Simulated parallel cost undertaking with three distinct cost distributions.

Main Results:

• The proposed algorithm effectively identifies feature subsets with maximal importance and minimal cost.
• It simultaneously selects a predetermined number of optimal features.
• Experimental results demonstrate the algorithm's efficiency and superiority over existing cost-sensitive methods.

Conclusions:

• The rough sets and Laplacian score based algorithm offers an efficient and effective approach to cost-sensitive feature selection.
• This method improves upon heuristic algorithms by preserving locality and considering feature interdependencies.
• The algorithm shows promising results for obtaining minimum cost feature subsets in various applications.