Feature Selection Based on Adaptive Particle Swarm Optimization with Leadership Learning.

Area of Science:

• Computer Science
• Artificial Intelligence
• Data Mining

Background:

• The Internet of Things (IoT) era presents challenges with high-dimensional data, making feature selection crucial.
• Particle Swarm Optimization (PSO) is effective for feature selection but struggles with local optima in high-dimensional spaces due to fixed parameters and low diversity.
• Existing metaheuristic algorithms often lack robust exploration and exploitation for complex datasets.

Purpose of the Study:

• To propose an improved feature selection method, adaptive Particle Swarm Optimization with Leadership Learning (APSOLL).
• To address the limitations of traditional PSO in handling high-dimensional datasets.
• To enhance the efficiency and effectiveness of feature selection in the context of IoT and big data.

Main Methods:

• Developed an adaptive PSO variant (APSOLL) incorporating an adaptive parameter updating strategy.
• Integrated a leadership learning mechanism to enhance population diversity and prevent local optima.
• Evaluated APSOLL on 10 UCI datasets against various optimization algorithms and traditional feature selection methods.

Main Results:

• APSOLL demonstrated superior exploration and exploitation capabilities compared to PSO, GWO, HHO, FPA, SSA, LPSO, and HPSO-DE.
• The method achieved significant dimensionality reduction, selecting less than 8% of original features on average.
• Generated feature subsets were more effective than those from ANOVA, CHI2, Pearson, Spearman, Kendall, and MI.

Conclusions:

• APSOLL offers a robust and efficient solution for feature selection in high-dimensional datasets.
• The adaptive strategies and leadership learning effectively overcome PSO's limitations.
• This approach holds promise for improving machine learning model performance in data-intensive applications.