Expert Systems最新文献

Dimensionality reduction plays a vital role in enhancing the performance of data classification tasks by reducing the complexity of the feature space. This study examines the effectiveness of integrating dimensionality reduction techniques with classification algorithms across four strategic configurations: (1) machine learning (ML)-based dimensionality reduction with ML classifiers, (2) deep learning (DL)-based dimensionality reduction with DL classifiers, and two heterogeneous combinations that mix ML and DL methods. Using 20 benchmark datasets from diverse domains, with feature dimensions ranging from 44 to 19,993, we systematically evaluate and compare these configurations. The dimensionality reduction methods include three ML-based feature selection techniques, Genetic Algorithm (GA), Information Gain (IG), and the C4.5 decision tree, and four DL-based feature extraction approaches, Autoencoder (AE), Sparse Autoencoder (SAE), Denoising Autoencoder (DAE), and Variational Autoencoder (VAE). For classification, Support Vector Machine (SVM) and k-Nearest Neighbours (KNN) are used as ML classifiers, while Multilayer Perceptron (MLP) and Deep Belief Network (DBN) serve as DL classifiers. Experimental results show that SAE consistently produces the most compact feature sets and improves classification performance, with the SAE + MLP combination achieving the best overall results. Furthermore, we explore ensemble dimensionality reduction strategies that integrate multiple algorithms. Although the best ensemble approach slightly outperforms the SAE + MLP model, the observed performance improvements are not statistically significant, that is, 0.839 versus 0.836 for AUC rates. In addition, SAE achieves a significantly higher dimensionality reduction rate compared to the best ensemble method (63% vs. 18%).

Many mobile devices—including digital cameras, smartphones, and personal digital assistants (PDAs)—rely on single image sensors to capture scenes for real-time processing. Convolutional neural networks (CNNs) have shown outstanding performance in various image processing tasks. In this paper, we propose a novel demosaicking method based on a transformer and long-range attention network (TLAN). The approach begins by initializing the mosaicked image using a bicubic interpolation algorithm, which provides a coarse reconstruction. TLAN is then applied to refine the output and accurately reconstruct the three colour channels. Our TLAN architecture combines the Swin Transformer (ST) with a dedicated long-range attention (LA) mechanism. The overall framework consists of both shallow and deep feature extraction modules. The deep extraction module is built from multiple residual swin transformer blocks (RSTBs), each composed of several Swin Transformer layers and augmented with a long-range attention block (LAB) to capture extended spatial dependencies. Experimental results demonstrate that the proposed method achieves superior performance in both PSNR and visual quality compared to existing demosaicking techniques.

Real coded genetic algorithms (RCGAs) are among the most versatile metaheuristic algorithms used to determine the global optimal solutions for nonlinear optimization problems, aimed at solving real-life complex optimization problems. After the selection operator, the crossover and mutation operators are two crucial strategies upon which the performance of a genetic algorithm (GA) depends. At present, various types of crossover and mutation operators are available in the literature. Therefore, this study attempts to present a state-of-the-art review on real coded crossover operators and real coded mutation operators. Each operator is explained with the help of examples. The objective of this state-of-the-art paper is to serve as a foundation for researchers who wish to design new real coded crossover operators or mutation operators. Moreover, to evaluate the effectiveness of various variants of RCGA, 23 classical benchmark problems are utilised, offering insights into their performance across different optimization problems.

This study sheds light on a fundamental problem in image segmentation known as multilevel image thresholding. With the rapid growth of artificial intelligence applications that rely on image processing such as medical imaging, remote sensing, and pattern recognition the demand for more effective techniques has become increasingly urgent. Traditional methods suffer from significant limitations, including slow convergence and premature convergence to local optima, particularly when applied to complex or high-dimensional images. To address these challenges, this study proposes a novel approach based on metaheuristic algorithms, specifically elephant herding optimization (EHO) and symbiotic organism search (SOS). Although these algorithms have shown promising results due to their adaptability and exploratory capabilities, they still face performance bottlenecks resulting from insufficient diversity in the search process. To overcome these limitations, enhanced variants of EHO and SOS are introduced by integrating opposition-based learning (OBL) and chaos theory to achieve a better balance between exploration and exploitation. These improved algorithms, OCEHO and OCSOS, are applied to the multilevel thresholding problem using Otsu's variance, Kapur's entropy and Masi's entropy as objective functions. The proposed methods are evaluated on 75 standard benchmark images, with segmentation quality evaluated using PSNR, SSIM, and FSIM metrics. Experimental results on 75 standard benchmark images show that the proposed OCEHO algorithm achieves PSNR values up to 37.51 dB, SSIM scores of 0.972, and FSIM values of 0.986, significantly outperforming baseline and hybrid variants. Furthermore, statistical analyzes, including the Wilcoxon rank sum test, confirm the superior stability and convergence speed of OCEHO over its counterparts. These results validate the effectiveness and robustness of the proposed approach for high-quality image segmentation.

Link prediction (LP) is a prominent research topic in network science and complex network analysis, focused on predicting future connections between unconnected node pairs by examining network topology and related characteristics. This paper introduces an advanced LP model combining the graph attention network (GAT) with Jaccard similarity and node centrality measures, such as local interaction density (LID) and hubs and authority centrality (HAC). Initially, a weight matrix incorporating structural and topological information is created using a feature matrix derived from similarity measures and node centrality. Then, node embeddings are generated using GAT, allowing the model to learn detailed representations that consider local and global contexts. GAT's multi-head attention mechanism enables the model to focus on various aspects of the node neighbourhood, capturing diverse structural information. A well-defined dataset is created from these embeddings, representing nodes at the endpoints of edges labelled as positive or negative. Linear discriminant analysis (LDA) is applied to this well-balanced and labelled dataset to perform classification tasks, leading to accurate link prediction. The model's performance is evaluated across eight different datasets, and the obtained results reveal the proposed model's superiority over several baseline and recently proposed LP models.