Abstract Data preprocessing is a pivotal stage in Machine Learning (ML) workflows, directly influencing both predictive performance and computational efficiency. Although widely acknowledged, the quantitative impact of preprocessing choices, particularly dimensionality reduction strategies, remains insufficiently explored in classification tasks. This study compares three data preparation approaches: raw data, Principal Component Analysis (PCA), and Principal Component Factor Analysis (PCFA). While PCA and Factor Analysis (FA) are well-established techniques, their structured integration as PCFA has received limited attention. By evaluating these alternatives, this study quantifies how dimensionality reduction affects predictive capability, result stability, and processing time across diverse learning scenarios. A factorial Design of Experiments (DOE) framework is employed, enabling the assessment of model performance under multiple attribute configurations and their interaction effects. Ten distinct ML algorithms are applied to ten heterogeneous datasets from different domains, with model training performed using cross-validation and hyperparameter optimization via Grid Search and Optuna. The findings show that PCA and PCFA consistently achieve competitive predictive performance while substantially reducing computational cost compared to raw data. Furthermore, data structure and attribute interactions strongly influence outcomes, indicating that the relationship between preprocessing strategies and model behavior is scenario-dependent. We conclude that combining dimensionality reduction, structured feature design, and optimized training strategies is essential for balancing predictive accuracy and computational cost, a trade-off increasingly critical in modern ML applications.

Abstract In order to solve challenging engineering problems, the state‐of‐the‐art in multi‐objective optimization shows a trend toward using meta‐heuristics and a posteriori decision‐making methods. This encourages the search for algorithms better able to find Pareto fronts with more convergence, coverage, and lower computational cost. This work shows the creation and validation of the Multi‐objective Sunflower Optimization (MOSFO), a hypercubic and constrained multi‐objective meta‐heuristic inspired by the phototropic life cycle of sunflowers around the sun. Having a much simpler programming model than most evolutionary algorithms, MOSFO was validated using the most difficult set of test functions in the literature (CEC 2009) and applied to ten constrained multi‐objective optimization problems (CEC 2021). The proposed algorithm was compared with ten other powerful algorithms: MOGWO, MOPSO, NSGA‐II, MOEA/D, NSGA‐III, CCMO, ARMOEA, ToP, TiGE 2, and AnD. Inverted General Distance, Spacing, Maximum Spread, and Hyper volume were used as comparison metrics to evaluate the convergence and coverage capabilities of the algorithms. MOSFO had the best average IGD value in 8 of the 10 test functions when compared with the other algorithms. In terms of MS, MOSFO had the highest average value of MS for 7 of the test functions. In summary, MOSFO showed substantial convergence and coverage capabilities and proved to be very competitive among the algorithms used, which were carefully selected to be popular and recent. The method is even more promising for problems with three or more objectives.

Abstract Feature selection is a knowledge discovery tool to understand the problem by analysing features. In particular, the application of feature selection in data mining can not only improve the quality of extracted patterns and knowledge but also decrease computational costs. Various techniques have been applied to this complex optimization problem, in which metaheuristics have been validated to be superior. This study introduces a new metaheuristic known for having lean and fast programming, inspired by the sunflower's motions for feature selection for the first time. It is equipped with a v‐shaped transfer function and associated with the KNN classifier to become the binary sunflower optimization (BSFO). A total of 12 variants of BSFO are designed based on the chaos theory and Lévy flights, called improved binary sunflower optimization (IBSFO). A discussion between these improvement theories for feature selection has also not been made yet, and it is performed in this paper using 15 benchmark datasets from the UCI repository. The experimental results show that all variants can advance the fitness value of BSFO, and nine of them considerably decrease the computational costs. Furthermore, the chaotic BSFO with the Chebyshev function, taking replacement to normal rand, has the lowest fitness value (−11.37%) and execution time (−9.31%) than the original BSFO. Further, IBSFO is compared with another eight metaheuristics and outperforms these competitors on average fitness value and execution time. Overall, IBSFO proved to find subsets with reduced dimension and high accuracy with meagre computational cost due to its robust explorative and exploitative capacities.