Advances in computational intelligence最新文献

This paper explores the integration of hesitant bipolar fuzzy soft sets (HBFSS) into multi-criteria group decision-making (MCGDM), aiming to enhance decision precision and resolve uncertainties in expert evaluations. We introduce a novel decision-making framework that combines the root mean square deviation (RMSD) method with a credibility score, capturing both the proximity to ideal solutions and the consistency of expert opinions. The process is applied to a sustainable energy project selection problem, showcasing its efficacy in ranking alternatives such as solar farm, wind park, and hydroelectric plant. A comparative analysis with the existing model highlights the limitations of traditional approaches, including the failure to differentiate alternatives with similar scores and neglecting expert consistency. Our results demonstrate that the proposed RMSD-Credibility approach offers a more nuanced, consistent, and precise ranking, improving decision quality in complex, uncertain environments. This paper contributes to advancing decision-making under fuzzy and uncertain conditions by providing an innovative aggregation mechanism tailored to the challenges of real-world multi-criteria problems.

Predicting personal traits, particularly age group, gender, handedness, and hand(s) used, in the form of digital identity for smartphone users by analysing keystroke dynamics (KD) attributes is a challenging area. However, it has a variety of applications in e-commerce, e-banking, e-teaching/learning, e-exams, forensics, and social networking. The main bottleneck of this problem is addressing the imbalanced nature of KD datasets using conventional machine learning (ML) approaches. By their inherent nature, KD datasets are often imbalanced from various perspectives due to the non-uniformity of diverse user traits and their varied usage patterns. This study proposes a predictive model for both fixed and free-text modes, considering the effect of attached smartphone sensors. We adopt a score-level fusion of eXtreme Gradient Boosting (XGBoost) models on several balanced bootstrapped training samples to address the limitations of conventional approaches. This ensemble approach utilizes multiple bootstrapped training sets, where the class distribution in each set is equally balanced for more accurate and robust performance. Furthermore, we observe the positive impact of incorporating these prediction scores and labels with primary biometric attributes in KD-based user authentication and identification, both in static/entry-point and continuous/active security designs—a previously unanswered challenges. The predictive mechanism and its adaptation in unique KD-based designs, based on datasets collected from a considerable number of volunteers with diverse age groups, genders, professions, and education levels through a smartphone in a web environment, demonstrate the novelty of our approach.

Keyword spotting (KWS) is a critical component of voice-driven smart-device applications, requiring high accuracy, sensitivity, and responsiveness to deliver optimal user experiences. Given the always-on nature of KWS systems, minimizing computational complexity and power consumption is essential, particularly for battery-powered edge devices with constrained resources. In this paper, we propose a compact and highly efficient convolutional neural network (CNN) for edge-based KWS tasks, using the Google Speech Commands (GSC) V2 dataset for training and evaluation. Our model employs modified MobileNetV2 architecture, optimized via knowledge distillation from an ensemble of high-performing CNN models. Experimental results demonstrate that the proposed model achieves 94.48% accuracy on clean test data and significantly outperforms existing state-of-the-art edge models on challenging noisy test sets, reaching 86.38% accuracy. The proposed CNN maintains this superior performance with only 73.8K parameters and 19.5M floating-point operations (FLOPs)—approximately three times fewer FLOPs and substantially fewer parameters than previously reported edge-focused KWS models. Moreover, when evaluated on a realistic and challenging external Kaggle test set, the proposed model shows excellent generalization with 88.38% accuracy, surpassing baseline depthwise separable CNN (DS-CNN) approaches. Upon practical deployment on a widely used embedded computing platform, our optimized model achieved fast inference times between 11 ms and 14 ms per sample, outperforming existing baseline methods and confirming its suitability for real-time applications. This study highlights the successful integration of model compression techniques, including ensemble learning and knowledge distillation, to achieve breakthrough performance improvements in accuracy, robustness to noise, computational efficiency, and inference speed, thereby advancing the practical deployment of high-performance KWS solutions on resource-constrained edge devices.

A meme is a part of media created to share an opinion or emotion across the internet. Due to their popularity, memes have become the new form of communication on social media. However, they are used in harmful ways such as trolling and cyberbullying progressively due to their nature. Various data modelling methods create different possibilities in feature extraction and turn them into beneficial information. The variety of modalities included in data plays a significant part in predicting the results. We try to explore the significance of visual features of images in classifying memes. Memes are a blend of both image and text, where the text is embedded into the picture. We consider a meme to be trolling if the meme in any way tries to troll a particular individual, group, or organisation. We try to incorporate the memes as a troll and non-trolling memes based on their images and text. We evaluate if there is any major significance of the visual features for identifying whether a meme is trolling or not. Our work illustrates different textual analysis methods and contrasting multimodal approaches ranging from simple merging to cross attention to utilising both worlds’—visual and textual features. The fine-tuned cross-lingual language model, XLM, performed the best in textual analysis, and the multimodal transformer performs the best in multimodal analysis.

Developing accurate and robust prognostic prediction for classifying the risks of acute lymphoblastic leukemia (ALL) relapse is critical for patient treatment management and survival. However, the lack of clinical samples and linearity assumption remains a significant clinical challenge for achieving high accuracy for single-cell prognostics. Here, we explore the use of non-linear machine learning models with ex vivo sample perturbation as a data augmentation strategy to improve ALL relapse prediction. We hypothesize that treating each sample with ex vivo perturbation can be viewed as independent measurements, thus increasing the number of available observations for machine learning. We show that ex vivo sample stimulation combined with non-linear machine learning significantly improves the performance of ALL risk stratification from limited single-cell proteomic data.

Optimization algorithms are continuously evolving and considered as an active multidiscipline research area to design scalable solutions for complex optimization problems. Literature witnesses the constant effort by researchers to improve existing optimization algorithms or to develop a new algorithm to deal with single and multiple objective problems. This research paper presents a novel population-based, metaheuristic bio-inspired optimization algorithm. The algorithm contrived the propagation concept of antibiotic-resistant bacteria named as antibiotic-resistant bacteria propagation (ARBP) algorithm where properties of bacteria to acquire antibiotic resistance over time are used as a base concept. The optimization algorithm imitates the two prime mechanisms of horizontal gene transfer—Conjugation Gene Transfer Mechanism (CGTM) and Transformation Gene Transfer Mechanism (TGTM) to propagate antibiotic-resistant bacteria. CGTM and TGTM are used to explore the search space to handle single and multiple objective optimization problems. Conjugation mechanism is used for exploration of search space and exploitation concept is driven by transformation mechanism. The efficiency and importance of the ARBP algorithm are validated on varying classical and complex benchmark functions. An extensive comparative study is performed to detail the effectiveness of ARBP over other well-known swarm and evolutionary algorithms. This comparative analysis clearly depicts that the performance of ARBP is superior in terms of finding a better solution with high convergence as compared to other considered algorithms.

Fundus images are a powerful tool for detecting a variety of retinal disorders. Regular screening of the retina can lead to early detection of conditions like diabetic retinopathy, allowing for timely intervention and treatment. This study is focussed on developing an automated diagnostic system that can accurately detect different stages of diabetic retinopathy. Our approach involves leveraging pre-trained deep learning system to extract important features from fundus images. These features are then employed in a classification system that categorises the images into five stages of retinopathy based on ensemble algorithms. We employ ensemble algorithms like Random forest and XGBoost for classification to improve the accuracy and predictability of the forecast. This drives our focus on enhancing the interpretability and explainability of the model. We trained the model using publicly available fundus images of diabetic individuals for grading and compared the classification results obtained from ensemble techniques with those from deep learning models that used pre-trained weights and biases. The best performing ensemble showed an accuracy range of 0.63 to 0.79. Moreover, the accuracy of 0.96 in detecting the presence of retinopathy provides strong evidence of the approach’s effectiveness, contributing to its reliability, and potential for early diagnosis.

During the last decade, the Chinese housing market has seen fast expansion, and the importance of housing price forecasts has surely increased, becoming an essential problem for policymakers and investors. In this article, we explore Gaussian process regressions across different kernels and basis functions for monthly office real estate price index forecasts for ten major Chinese cities from July 2005 to April 2021 using cross-validation and Bayesian optimizations that could endow the forecast models with higher adaptability and better generalization performance. The models constructed offer precise out-of-sample forecasts from May 2019 to April 2021 with relative root mean square errors ranging from 0.0205 to 0.5300% across the ten price indices. Benchmark analysis against the autoregressive model, autoregressive-generalized autoregressive conditional heteroskedasticity model, nonlinear autoregressive neural network model, support vector regression model, and regression tree model suggests that the Gaussian process regression model leads to statistically significant higher accuracy. Our findings might be utilized independently or in conjunction with other projections to create views on office real estate price index movements and undertake further policy research.

An approach to improve workers’ productivity performance without neglecting their well-being should be investigated. To elucidate the effects of systematic micro-break on intellectual concentration performance, a controlled laboratory experiment generated 31 participants’ data when each participant was performing cognitive comparison tasks. Systematic micro-break was given for 20 s after 7.5 min of cognitive work, for a total of 25 min of work tasks. Each participant performed the task under both conditions with and without micro-break intervention in a counterbalanced design. Two quantitative evaluations were made: the answering time and concentration time ratio. A subjective symptom questionnaire and the NASA task load index were applied for analytical consideration. The average answering time indicates that the performance under the influence of micro-break tends to be more stable over time and that it mitigates performance degradation compared to the performance in a condition without micro-break. For concentration time ratio scores, no significant difference was found between conditions with micro-break and without micro-break. However, a tendency was apparent by which the concentration time ratio score was higher in a condition with micro-break, which suggests higher cognitive performance. The subjective symptoms questionnaire indicated no significant difference between conditions with and without micro-break. Weighted NASA task load index questionnaire results indicated significant difference between both conditions with lower workload scores in conditions with micro-break. Results obtained from this study suggest that the implementation of systematic micro-break can support workers’ performance stability over time. Therefore, systematic micro-break can be promoted as a promising strategy for work recovery.

Colour differentiation is crucial in machine learning and computer vision. It is often used when identifying items and objects based on distinct colours. While common colours like blue, red, green, and yellow are easily distinguishable, some applications require recognising subtle colour variations. Such demands arise in sectors like agriculture, printing, healthcare, and packaging. This research employs prevalent unsupervised learning techniques to detect printed colours on paper, focusing on CMYK ink (saturation) levels necessary for recognition against a white background. The aim is to assess whether unsupervised clustering can identify colours within QR-Codes. One use-case for this research is usage of functional inks, ones that change colour based on environmental factors. Within QR-Codes they serve as low-cost IoT sensors. Results of this research indicate that K-means, C-means, Gaussian Mixture Model (GMM), Hierarchical clustering, and Spectral clustering perform well in recognising colour differences when CMYK saturation is 20% or higher in at least one channel. K-means stands out when saturation drops below 10%, although its accuracy diminishes significantly, especially for yellow or magenta channels. A saturation of at least 10% in one CMYK channel is needed for reliable colour detection using unsupervised learning. To handle ink densities below 5%, further research or alternative unsupervised methods may be necessary.