A unified feature and instance selection framework using optimum experimental design.

Area of Science:

• Machine Learning
• Data Science
• Statistical Modeling

Background:

• Feature selection and active learning are crucial for efficient data analysis but are typically addressed independently.
• These processes are inherently dual operations on data matrices, suggesting potential for a unified approach.

Purpose of the Study:

• To introduce a novel method for simultaneously selecting informative features and instances.
• To leverage optimum experimental design principles for minimizing parameter estimation variance.

Main Methods:

• Propose the Unified criterion for Feature and Instance selection (UFI) method.
• Formulate the problem as minimizing the trace of the parameter covariance matrix.
• Employ a greedy algorithm for efficient optimization.

Main Results:

• The UFI method effectively identifies optimal feature and instance subsets.
• Experimental results on benchmark datasets validate the proposed approach.

Conclusions:

• Simultaneous feature and instance selection offers a more effective strategy than independent approaches.
• The UFI method provides a robust framework for enhancing machine learning model performance.