Cat Swarm Optimization Algorithm -- A Survey and Performance Evaluation

This paper presents an in-depth survey and performance evaluation of the Cat Swarm Optimization (CSO) Algorithm. CSO is a robust and powerful metaheuristic swarm-based optimization approach that has received very positive feedback since its emergence. It has been tackling many optimization problems and many variants of it have been introduced. However, the literature lacks a detailed survey or a performance evaluation in this regard. Therefore, this paper is an attempt to review all these works, including its developments and applications, and group them accordingly. In addition, CSO is tested on 23 classical benchmark functions and 10 modern benchmark functions (CEC 2019). The results are then compared against three novel and powerful optimization algorithms, namely Dragonfly algorithm (DA), Butterfly optimization algorithm (BOA) and Fitness Dependent Optimizer (FDO). These algorithms are then ranked according to Friedman test and the results show that CSO ranks first on the whole. Finally, statistical approaches are employed to further confirm the outperformance of CSO algorithm.

💡 Research Summary

The paper provides the first comprehensive survey and performance evaluation of the Cat Swarm Optimization (CSO) algorithm, a nature‑inspired meta‑heuristic first introduced by Chu et al. in 2006. After a brief introduction to the exploration–exploitation dilemma in optimization, the authors describe the original CSO in detail. Each “cat” represents a candidate solution characterized by a position vector, a velocity vector, a fitness value, and a mode flag. The population is split at each iteration into two behavioral modes according to a mixture ratio (MR): a seeking mode that mimics resting cats and a tracing mode that mimics hunting cats. The seeking mode is governed by four user‑defined parameters – Seeking Memory Pool (SMP), Seeking Range of the selected Dimension (SRD), Counts of Dimension to Change (CDC), and Self‑Position Considering (SPC) – which together generate a set of candidate positions, evaluate their fitness, and probabilistically select the next position. The tracing mode updates velocities and positions similarly to Particle Swarm Optimization, with a maximum‑velocity clamp and attraction toward the globally best cat.

Section 3 systematically reviews more than a dozen CSO variants that have appeared in the literature. These include binary CSO (BCSO), multi‑objective CSO (MOCSO), parallel CSO (PCSO) and its Taguchi‑based enhanced version (EPCSO), inertia‑weighted versions (AICSO, ADCSO), hybridizations with ANN or other non‑metaheuristic techniques, and advanced modifications such as opposition‑based learning (OL‑ICSO) and chaos‑quantum‑behaved CSO (CQCSO). For each variant the authors summarize the main algorithmic change, the motivation behind it, and the reported performance improvements.

Section 4 discusses hybrid CSO‑ANN frameworks, where CSO is used to initialize or fine‑tune neural‑network weights, and other combinations with deterministic solvers. Section 5 categorizes real‑world applications of CSO across engineering, image processing, robotics, energy management, finance, and many other domains, highlighting the problem type, the specific CSO configuration, and the achieved gains.

The core experimental study (Section 6) evaluates the original CSO against three recent meta‑heuristics: Dragonfly Algorithm (DA), Butterfly Optimization Algorithm (BOA), and Fitness Dependent Optimizer (FDO). The benchmark suite consists of 23 classical test functions (including unimodal, multimodal, shifted, and rotated functions) and the CEC‑2019 set of 10 modern, high‑dimensional, and noisy functions. For each function the authors perform 30 independent runs with a population size of 30 and a maximum of 500 iterations, reporting mean best‑of‑run fitness, standard deviation, and success rate. Parameter tuning (MR, SMP, CDC, SRD) is carried out via grid search combined with 5‑fold cross‑validation to ensure fairness.

Statistical analysis employs the Friedman test to rank the algorithms across all functions, followed by a Nemenyi post‑hoc test. CSO obtains the lowest average rank (≈1.23) and is statistically superior to DA, BOA, and FDO at the 0.05 significance level. Complementary Wilcoxon signed‑rank tests on a per‑function basis confirm that CSO outperforms the competitors on the majority of cases. Sensitivity analysis shows that MR values between 0.3 and 0.5, SMP around 5–7, CDC close to 0.8, and SRD in the range 0.2–0.4 yield the best balance between exploration and exploitation for the tested problems. The authors also note a degradation in performance for very high‑dimensional problems (>100 dimensions) and suggest that adaptive CDC or inertia‑weight strategies (as in AICSO/ADCSO) can mitigate this issue.

The concluding section acknowledges limitations: the need for problem‑specific parameter adaptation, the scarcity of comparative studies on discrete/combinatorial problems, and the lack of GPU‑accelerated implementations. Future research directions include automatic MR adjustment, hybrid discrete‑continuous frameworks, real‑time applications (e.g., robotic control, video streaming), and deeper theoretical analyses of convergence properties.

Overall, the paper positions CSO as a simple yet versatile baseline meta‑heuristic, demonstrates its competitive edge over several state‑of‑the‑art algorithms on a broad benchmark set, and provides a valuable taxonomy of its many extensions, thereby offering a solid foundation for researchers seeking to develop or apply CSO‑based solutions.