SMCFO: a novel cuttlefish optimization algorithm enhanced by simplex method for data clustering.

Area of Science:

• Data Science
• Machine Learning
• Optimization Algorithms

Background:

• Unsupervised learning relies on data clustering, but existing algorithms like Kmeans and Cuttlefish Optimization Algorithm (CFO) face challenges such as premature convergence and poor local search.
• These limitations hinder effective processing of complex or imbalanced datasets and handling non-spherical cluster shapes.

Purpose of the Study:

• To introduce a novel clustering algorithm, SMCFO, that enhances the Cuttlefish Optimization Algorithm (CFO) by integrating the Nelder-Mead simplex method.
• To address the limitations of existing clustering algorithms, including premature convergence and inadequate local search capabilities.

Main Methods:

• The proposed SMCFO algorithm partitions the population into four subgroups, each with distinct update strategies.
• One subgroup employs the Nelder-Mead method for solution quality enhancement, while others balance exploration and exploitation.
• Performance was evaluated against CFO, PSO, SSO, and SMSHO using 14 datasets, including artificial and benchmark datasets from the UCI Machine Learning Repository.

Main Results:

• SMCFO consistently outperformed all comparison algorithms across all tested datasets in terms of clustering accuracy, convergence speed, and stability.
• Nonparametric statistical tests confirmed the statistically significant superiority of SMCFO's performance.
• The simplex-enhanced design was identified as the key factor boosting local exploitation and stabilizing convergence.

Conclusions:

• The SMCFO algorithm represents a significant advancement in data clustering, offering improved performance over existing metaheuristic and traditional methods.
• The integration of the Nelder-Mead method effectively enhances local search capabilities and stabilizes the optimization process.
• SMCFO demonstrates robust and statistically significant improvements, making it a promising tool for complex clustering tasks.