This paper proposes Adaptive Ensemble Multi-Objective Differential Evolution for Feature Selection (AEMO-DEFS), a wrapper-based framework that simultaneously minimizes feature count and maximizes classification accuracy for high-dimensional gene expression data. AEMO-DEFS introduces two key adaptive mechanisms: (1) an ensemble of DE mutation strategies (rand/1/bin, best/1/bin, current-to-pbest/1/bin) with dynamic selection based on historical success, and (2) memory-based self-adaptation of the scaling factor F (using Cauchy distribution) and crossover rate CR (using Normal distribution). Non-dominated sorting combined with crowding distance maintains diverse Pareto-optimal solutions. Comprehensive experiments on five cancer microarray datasets demonstrate that AEMO-DEFS outperforms MOPSO, NSGA-III, and MOEA/D across all evaluation metrics, achieving superior Inverted Generational Distance (IGD), ϵ-indicator, classification accuracy, and significant feature reduction.

Diabetic retinopathy (DR) remains a leading cause of vision impairment worldwide, necessitating early detection to prevent irreversible blindness. Manual screening processes are time-consuming and constrained by resource limitations, emphasizing the need for automated and efficient diagnostic systems. Deep learning models have delivered notable results, but further refinement is still possible. In this work, we provide an advanced ensemble deep learning framework for automated DR screening, leveraging the Choquet fuzzy integral to aggregate the outputs of multiple state-of-the-art convolutional neural networks (CNNs), including DenseNet121, Xception, and InceptionResNetV2. These pretrained models are fine-tuned on retinal fundus images to extract complementary features, which are then dynamically combined using the Choquet fuzzy integral, thereby improving predictive accuracy and robustness. Our model was trained and evaluated on a comprehensive real-world dataset, outperforming conventional deep learning and traditional machine learning baselines in terms of accuracy, sensitivity, and specificity. Experiments demonstrated the model’s resilience to dataset variability and its strong generalization capabilities. The proposed approach attained an average recall of 84.09 %, a precision of 83.09 %, an F1-score of 82.96 %, and an accuracy of 84.09 %. These findings substantiate the potential of ensemble learning techniques in advancing the effectiveness of automated DR screening systems.

In this study, we propose an advanced Orthogonal Genetic Algorithm (OGA) specifically developed to tackle the Traveling Salesman Problem with Drones (TSP-D), a multifaceted optimization challenge that necessitates precise synchronization between a truck and a drone for effective delivery tasks. The OGA integrates Orthogonal Crossover and Region-Based Mutation strategies, thereby enhancing the algorithm's proficiency in optimizing drone routing in a range of TSP-D scenarios. This novel approach significantly augments the algorithm's adaptability and exploratory capabilities within the intricate search space. Our comprehensive experimental analysis rigorously evaluates the performance of the proposed OGA against established algorithms in a variety of TSP-D instances. The results from these evaluations reveal that our approach substantially surpasses conventional algorithms in terms of both convergence speed and solution quality. This enhanced performance underscores the OGA's efficacy and robustness in optimizing complex paths in TSP-D scenarios.

Breast cancer remains a significant global health concern, necessitating the advancement of image segmentation techniques to improve diagnostic accuracy. Traditional thresholding methods often fail to effectively segment images due to complex cellular structures and indistinct boundaries. To address these challenges, this study proposes the Local Homogeneity-Based Reinitialization Particle Swarm Optimization (LHBRPSO) algorithm for multi-threshold segmentation, integrating local homogeneity analysis with adaptive threshold reinitialization to enhance segmentation accuracy by mitigating intra-class variability. The performance of LHBR-PSO is evaluated using peak signal-to-noise ratio (PSNR) and structural similarity index measure (SSIM), ensuring a comprehensive assessment of both numerical accuracy and perceptual quality. Comparative analysis against conventional optimization-based segmentation methods demonstrates the superior performance of LHBR-PSO across various thresholding levels, while statistical validation using the Friedman ranking test further confirms its robustness, as it consistently achieves higher rankings in segmentation effectiveness. These findings underscore the potential of LHBR-PSO as a reliable and computationally efficient approach for image segmentation, offering an advanced solution for enhancing breast cancer diagnosis.

Software Defect Prediction (SDP) aims to detect bugs at an early stage of the software development process, helping to improve quality while reducing costs and development time. Machine learning (ML) models, such as LightGBM, have shown strong performance, their effectiveness depends heavily on proper hyperparameter optimization. This paper introduces MS-FSOA, a multi-strategy enhancement of the Starfsh Optimization Algorithm, to optimize LightGBM for SDP. MS-FSOA integrates enhanced population initialization, Lévy flight, and adaptable cooperative hunting to improve search quality and maintain diversity throughout the optimization process. Experiments on three PROMISE datasets show that MS-FSOA significantly improves prediction performance. Compared to traditional algorithms, it boosts LightGBM’s accuracy and robustness in software defect classification.

This paper introduces Dynamic Social Particle Swarm Optimization (DS-PSO), a novel adaptation of the traditional Particle Swarm Optimization (PSO) technique specifically engineered for complex optimization challenges. DS-PSO innovatively incorporates dynamic social interactions within the swarm, enhancing adaptability and addressing the typical limitations of premature convergence and limited exploration in conventional PSO. A key feature of DS-PSO is its ability to balance exploration and exploitation efficiently, making it particularly suitable for dynamic environments. The primary application highlighted in this study is automatic clustering, a crucial task in data analysis involving unsupervised data grouping without prior knowledge of cluster numbers. DS-PSO’s flexibility and improved search capability demonstrate its potential as an effective tool for automatic clustering, promising significant advancements in data-driven optimization and analysis.

This study proposes a new approach for the Facial Expression Recognition (FER) system that combines the Quantum Gazelle Optimization Algorithm (QGOA), local binary patterns (LBP), and Histograms of Oriented Gradients (HOG) with an optimized Deep Neural Network (DNN) classifier. The system uses computer vision techniques and deep learning algorithms to identify emotions in facial expressions. The proposed technique first employs the HOG and LBP descriptors, crucial components with excellent pattern recognition capabilities. These descriptors provide features resilient to small local changes in posture and lighting. However, they also generate unimportant and obtrusive characteristics that hinder classification performance. The proposed approach uses a wrapper-based feature selector called QGOA to solve this issue, which decreases the training complexity and improves recognition performance. QGOA takes advantage of the properties of quantum computing to regulate the diversity of face features and make proper selections using quantum measurements and Q-bit superstitious states. Finally, the optimized DNN detects facial emotions based on the selected features. The proposed approach was tested on the widely adopted FER2013 dataset. The results of the extensive analysis demonstrate the effectiveness of the proposed approach over state-of-the-art systems.

This paper presents a novel enhancement to Particle Swarm Optimization (PSO) by integrating parallax learning to improve convergence speed and solution quality. Comparative simulations against state-of-the-art metaheuristics on CEC 2022 demonstrate superior performance. Additionally, we apply our method to optimize Electric Vehicle (EV) Fast Charging Station (FCS) placement, achieving efficient solutions considering travel time and power loss costs on an IEEE 33-bus test system. Our algorithm not only surpasses competitors on standard benchmarks but it also provides high-quality solutions to realistic optimization problems in energy systems planning and operation.

This paper introduces a supply chain quality sustainability decision support system (QSDSS). It uses association rule techniques to provide better logistics plans and handle risk in the supply chain. An improved Multiobjective Crystal Structure Algorithm using centroid opposition based learning and gaussian perturbation is proposed to avoid premature convergence and escape from local optimal. The experimental study is carried out in two phases, using two classic benchmark suites and the “DataCo SMART SUPPLY CHAIN” dataset for big data analysis.

In cluster analysis, classical approaches suffer from the problem of identifying the number of clusters, known as the automatic clustering problem. Therefore, automatic clustering has become a popular research area and offers opportunities in various data analysis applications such as bioinformatics, medicine, image processing and consumer segmentation. It is considered as NP- complete problem where it is preferable to use approximate approaches. In this study, we propose an hybrid approach between chaos game optimization and particle swarm optimization (CGOPSO). The Davies-Bouldin index (DBI) is used as a main objective of the proposed approach with the purpose to find the most accurate number of cluster centroids and their positions. To assess its performance, we compared CGOPSO with different other existing algorithms in the literature over 12 classical datasets using two different validity indexes: Davies Bouldin index (DBI) and Compact-Seperated index (CSI). The experimental results have demonstrated that CGOPSO shows better performance than other algorithms.

To obtain computer procedures that intelligently guide the search process by efficiently exploring the search space corresponds to an optimization problem that is solved using meta-heuristics. However, there are many optimization problems whose objective is to extract a single efficient solution that reflects the quality of the system performance. Hence, we also encounter multi-objective problems. This paper studies a multi-objective optimization problem that addresses the complexity of problems with synchronous indices. In this paper, association rule mining (ARM) is treated as an optimization problem. In recent literature, most of the proposed methods of ARM generate a large number of redundant and irrelevant rules. Therefore, we propose an improved multi-objective crystal structure algorithm using the TOPSIS approach. The experimental study is performed in two steps. First, we use a classical benchmark suite, namely WFG, to analyze the effectiveness of the proposed approach in generating solutions close to the Pareto fronts. In a second step, to confirm the performance of our algorithm, we apply it on the database of the Bosch Production line performance, and we conduct a comparative evaluation towards recent algorithms. The obtained results demonstrate the efficiency of the proposed algorithm in terms of number of rules, average support, average confidence, average conviction, average certain factor.

The jellyfish search optimizer (JSO) is one of the newest swarm intelligence algorithms which has been widely used to solve different real-world optimization problems. However, its most challenging task is to regulate the exploration and exploitation search to avoid problems in harmonic convergence or be trapped into local optima. In this paper, we propose a new variant of JSO named OJSO, based on orthogonal learning with the aim to improve the capability of global searching of the original algorithm. The orthogonal learning is a strategy for discovering more useful information from two recent solution vectors by predicting the best combination using limited trials instead of exhaustive trials via an orthogonal experimental design. To evaluate the effectiveness of our approach, 23 benchmark functions are used. The evaluation process leads us to conclude that the proposed algorithm strongly outperforms the original algorithm in all aspects except the execution time.

Feature selection (FS) is considered a challenging machine learning problem that handles the large size of features. The main purpose of FS is to remove irrelevant and redundant variables to improve the performance of the learning algorithms. Consequently, the FS process is considered as an optimization problem where metaheuristics approaches prove efficiency in solving it. In this paper, we propose new binary versions of a new ancient inspired metaheuristic approach called Giza Pyramids Construction (GPC) to select the most relevant subset of features. The proposed binary versions of the algorithm called BGPC-S and BGPC-V are implemented with two transfer functions, with the main objective of maximizing classification accuracy and minimizing the number of selected features. The two versions of BGPC were compared to six well-known binary metaheuristics for feature selection problem, namely Binary Atom Search Optimisation (BASO), Binary Bat Algorithm (BBA), Binary Differential Evolution (BDE), Binary Grey Wolf Optimizer (BGWO), Binary Particle Swarm Optimization (BPSO), and Binary Harris Hawks Optimizer (BHHO), and evaluated over 20 datasets from the UCI repository. Experiments have demonstrated that the proposed approaches outperformed the other algorithms in terms of classification accuracy and the number of selected features.

One of the great aspirations of machine learning is the clustering methods. It consists on categorized a set of similar data into different groups based on related properties. The clustering ensemble is used in aim to improve the performance and the stability of the unsupervised classification methods through the concept of weighting. One of the major problems in clustering ensembles is the consensus function. In this paper, we study the amalgamation of clustering techniques, trying to benefit from the strengths of each algorithm and we emerge the problem of combining multiple clustering of a set of objects. A new efficient for Consensus Functions of Cluster Ensembles is proposed based on Multi-layer networks technique. Experiments are carried out on a variety of datasets which highlights our proposed method.