A scalable memetic algorithm for simultaneous instance and feature selection.

Area of Science:

• Machine Learning
• Data Mining
• Pattern Recognition

Background:

• Increasing data volumes necessitate efficient data handling techniques.
• High-dimensional datasets pose significant challenges for classification and recognition tasks.
• Existing methods often address instance and feature selection independently, missing synergistic benefits.

Purpose of the Study:

• To propose a novel memetic algorithm for simultaneous instance and feature selection.
• To address the challenges of large datasets with many features and instances.
• To improve classification accuracy and algorithm efficiency.

Main Methods:

• A new memetic algorithm is developed for joint instance and feature selection.
• Four distinct local search procedures are employed to identify optimal instance and feature subsets.
• A novel fitness function is introduced to balance instance selection and feature reduction.
• An extended stratification approach enhances scalability for large datasets.

Main Results:

• Experimental validation demonstrates improved classification accuracy by the proposed fitness function.
• The algorithm shows effectiveness on 55 medium to large datasets from the UCI Machine Learning Repository.
• Successful application to 30 large-scale problems and 40 class-imbalanced datasets is reported.
• The method's utility is confirmed with decision trees and support vector machines.

Conclusions:

• The proposed memetic algorithm offers an effective solution for joint instance and feature selection.
• The method demonstrates strong performance, scalability, and accuracy on diverse and challenging datasets.
• This approach provides a valuable tool for tackling big data problems in pattern recognition.