ST-AL: a hybridized search based metaheuristic computational algorithm towards optimization of high dimensional...

Area of Science:

• Data Science
• Artificial Intelligence
• Optimization Algorithms

Background:

• High-dimensional data mining faces challenges with irrelevant and redundant features, impacting industrial process accuracy and efficiency.
• Meta-heuristic optimization algorithms are used for feature selection but struggle with global optimization challenges.
• Existing methods often fall into local optima, limiting their effectiveness in complex industrial data mining.

Purpose of the Study:

• To propose an improved Sooty Tern Optimization (ST) algorithm, named ST-AL, for enhanced search performance in high-dimensional industrial optimization problems.
• To address the limitations of traditional optimization algorithms in feature selection by improving exploration-exploitation balance and avoiding local minima.
• To enhance the accuracy and efficiency of industrial data mining through a novel feature selection approach.

Main Methods:

• The ST-AL method enhances the Sooty Tern Optimization Algorithm (STOA) using four key strategies: controlled randomization for exploration-exploitation balance, an Ant Lion (AL) based exploration phase, modified exploitation phase equations, and greedy selection to discard poor populations.
• The algorithm was evaluated on ten CEC2020 benchmark functions for global optimization.
• Performance was further assessed using feature selection on 16 UCI benchmark datasets, comparing against seven established optimization-based feature selection methods.

Main Results:

• The ST-AL algorithm demonstrated superior performance in avoiding local minima and accelerating convergence compared to existing methods.
• Experimental results showed a mean accuracy ranging from 0.94 to 1.00 when compared with state-of-the-art algorithms including ALO, STOA, PSO, GWO, HHO, MFO, and MPA.
• The proposed method effectively handles high-dimensional data, improving the efficiency and accuracy of feature selection.

Conclusions:

• The ST-AL method offers a significant improvement over existing optimization algorithms for high-dimensional feature selection.
• The enhanced algorithm effectively balances exploration and exploitation, leading to better global search capabilities and avoidance of local optima.
• The ST-AL algorithm presents a robust and accurate solution for complex industrial data mining challenges.