A Novel Adaptive Sand Cat Swarm Optimization Algorithm for Feature Selection and Global Optimization

Biomimetics, Journal Year: 2024, Volume and Issue: 9(11), P. 701 - 701

Published: Nov. 15, 2024

Feature selection (FS) constitutes a critical stage within the realms of machine learning and data mining, with objective eliminating irrelevant features while guaranteeing model accuracy. Nevertheless, in datasets featuring multitude features, choosing optimal feature poses significant challenge. This study presents an enhanced Sand Cat Swarm Optimization algorithm (MSCSO) to improve process, augmenting algorithm's global search capacity convergence rate via multiple innovative strategies. Specifically, this devised logistic chaotic mapping lens imaging reverse approaches for population initialization enhance diversity; balanced exploration local development capabilities through nonlinear parameter processing; introduced Weibull flight strategy triangular parade optimize individual position updates. Additionally, Gaussian-Cauchy mutation was employed ability overcome optima. The experimental results demonstrate that MSCSO performs well on 65.2% test functions CEC2005 benchmark test; 15 UCI, achieved best average fitness 93.3% fewest selections 86.7% attaining accuracy across 100% datasets, significantly outperforming other comparative algorithms.

Language: Английский

---

Machine Learning-Based DoS Amplification Attack Detection against Constrained Application Protocol

Applied Sciences, Journal Year: 2023, Volume and Issue: 13(13), P. 7391 - 7391

Published: June 21, 2023

This paper discusses the Internet of Things (IoT) and security challenges associated with it. IoT is a network interconnected devices that share information. However, low power resources make them vulnerable to attacks. Using heavy protocols like HTTP for can prove costly using popular lightweight CoAP invite attacks such as DoS (Denial-of-Service). While models DTLS LSPWSN secure against attacks, they also have limitations. To overcome this problem, proposes machine learning model detects amplification 99% accuracy. best our knowledge, research first use multi-classification process detect classify different types techniques attack client victim clients.

Language: Английский

---

A Hybrid Extreme Gradient Boosting and Long Short-Term Memory Algorithm for Cyber Threats Detection

MENDEL, Journal Year: 2023, Volume and Issue: 29(2), P. 307 - 322

Published: Dec. 19, 2023

The vast amounts of data, lack scalability, and low detection rates traditional intrusion technologies make it impossible to keep up with evolving increasingly sophisticated cyber threats. Therefore, there is an urgent need detect stop threats early. Deep Learning has greatly improved due its ability self-learn extract highly accurate features. In this paper, a Hybrid XG Boosted Long Short-Term Memory algorithm (HXGBLSTM) proposed. A comparative analysis conducted between the computational performance six established evolutionary computation algorithms recently developed bio-inspired metaheuristic called Zebra Optimisation Algorithm. These include Particle Swarm Algorithm, Bio-inspired Algorithms, Bat Firefly Monarch Butterfly as well Genetic Algorithm Evolutionary dimensionality curse been mitigated by using these methods for feature selection, results are compared wrapper-based selection XGBoost algorithm. proposed uses CSE-CIC -IDS2018 dataset, which contains latest network attacks. outperformed other FS was used evaluating effectiveness newly HXGBLSTM, binary multi-class classifications considered. When comparing HXGBLSTM threat detection, outperforms seven innovative deep learning classification four them classification. Other evaluation criteria such recall, F1 score, precision have also comparison. showed that best accuracy 99.8\%, F1-score 99.83\%, 99.85\%, recall 99.82\%, in extensive detailed experiments on real dataset. accuracy, F1-score, precision, were all around 100\%, does give advantage over ones.

Language: Английский

---

TensorCRO: A TensorFlow‐based implementation of a multi‐method ensemble for optimization

Expert Systems, Journal Year: 2024, Volume and Issue: unknown

Published: Sept. 5, 2024

Abstract This paper presents a novel implementation of the Coral Reef Optimization with Substrate Layers (CRO‐SL) algorithm. Our approach, which we call TensorCRO, takes advantage TensorFlow framework to represent CRO‐SL as series tensor operations, allowing it run on GPU and search for solutions in faster more efficient way. We evaluate performance proposed across wide range benchmark functions commonly used optimization research (such Rastrigin, Rosenbrock, Ackley, Griewank functions), show that execution leads considerable speedups when compared its CPU counterpart. Then, comparing TensorCRO other state‐of‐the‐art algorithms Genetic Algorithm, Simulated Annealing, Particle Swarm Optimization), results can achieve better convergence rates than within fixed time, given fitness are also implemented TensorFlow. Furthermore, approach real‐world problem optimizing power production wind farms by selecting locations turbines; every evaluated scenario, outperformed meta‐heuristics achieved close best known literature. Overall, our algorithm provides new, fast, solving problems, believe has significant potential be applied various domains, such engineering, finance, machine learning, where is often solve complex problems. propose this optimize models cannot propagate an error gradient, excellent choice non‐gradient‐based optimizers.

Language: Английский

---

Special issue on International conference on computing and communication networks (ICCCN2022)

Expert Systems, Journal Year: 2023, Volume and Issue: 41(7)

Published: Nov. 2, 2023

The past years have seen a flurry of activity in distributed computing and its related fields. Indeed, developments computer communications networks enabled the deployment exciting new areas, including internet things, vehicular networks, collaborative big data analysis so on. design implementation energy efficient future generation communication networking technologies fostered development mobile, pervasive large-scale technologies. International Conference on Computing Communications Networks (ICCCN 2022) aims to serve as forum for exchanging latest findings experiences ranging from theoretical research practical system all aspects networking. All submitted papers must substantial knowledge engineering component like AI/machine learning, etc. be scope this special issue, otherwise paper will get rejected straightaway. This issue brings selected presented at ICCCN 2022 conference. From around 45 articles, only 13 were based reviews. Each was reviewed by least two reviewers went through rounds brief contributions these are discussed below. In first authors Minni Jain et al. (2023) proposed approach code-mixed Hindi–English social media text that comprises language identification, detection correction non-word (out vocabulary) errors well real-word occurring simultaneously. A fuzzy graph between different words suggestive lists is generated using various semantic relations Hindi WordNet. Word embeddings graph-based centrality measures used find correct word. Several experiments performed datasets taken Instagram, Twitter, YouTube comments, Blogs WhatsApp. experimental results demonstrate corrects out-of-vocabulary with maximum recall 0.90 0.67, respectively, Dev_Hindi 0.87 0.66, Rom_Hindi. Ritu Bibyan their study prediction model LDA content aspect emotion sentiment aspect. validated collected Eclipse project convolutional neural network (CNN). show CNN effectively utilizes handle severity prediction. It also improves generalization overfitting avoided. Anshika Arora, Chakraborty, (2023), present novel methodology assessment objective sleep quality actigraph recordings motor activity. High level features sequential extracted Long-Short Term Memory (LSTM) model, which then paired significant statistical feature, namely zero percent, describes percentage events over series. predictive ability combined feature vector evaluated support machine (SVM) classifier. hybrid LSTM-SVM framework benchmark dataset, namely, MESA Actigraphy achieves an accuracy 85.62% concept overlapping sliding window performance 3.51%, use discriminative task 2.95%. Comparison state art validates indicator via actigraph-based data. next paper, Dimple Sethi gait markerless environment divided into three stages: First, database videos MNIT RAMAN LABORA-TORY Jaipur has been prepared. Second, skeletal landmark generate kinematic characteristics comparable gold standard marker-based techniques adapted. Third, MF-GA (multi-feature analysis), ensemble + LSTM, classification, developed. indicate can successfully low-cost clinical monitoring constraint-free environment. Our quantifying improve access quantitative clinics rehabilitation centres enable researchers conduct studies gait-related disorders. another research, Awanish Kumar Mishra introduced automated Cervical Precancerous Lesion Classification Quantum Invasive Weed Optimization Deep Learning (CPLC-QIWODL) biomedical pap smear images. CPLC-QIWODL technique examines images cervical cancer classification. Moreover, uses deep network-based SqueezeNet extraction. Furthermore, hyperparameter tuning takes place QIWO technique, showing novelty work. Finally, classify CC, variational autoencoder (DVAE) applied. result tested other existing algorithms, medical image database. Extensive comparative demonstrated enhanced outcomes 99.07%. Qiming Zhang marine radar internet-of-things make full echo AIS information target detection, radars work both scanning staring modes. By adopting segmentation learning methods, accurate perception weak plan-position (PPI) case X-band shows achieve high identification accuracy. Minzhi He studied application computed tomography (CT) intelligent algorithm recovery after surgery (ERAS) patients laparoscopic-assisted distal gastrectomy. An established collect lymph node metastasis data, groups compared. showed 54 cases predicted images, actual incidence 65.12%. PPS Bedi employed NLP-based algorithms summary perform linguistic summarization modify/adapt domain-specific summarization. provides developed in-house condensing ill-punctuated or unpunctuated discussion transcripts more intelligible summaries, combines topic modelling phrase selection punctuation restorations. For autonomous synthesis reports transcripts, proposes end-to-end Dense Long Short Network (LSTM), followed reveal models trained ordinary provide testing set, one outperforming test set. emphasizes popular soft computing, meta-heuristic named genetic (GA) solve optimization problem software system. Rajat Tandon, presents solution GA tool MATLAB. obtained relevance industry finding optimal values variables before release newly products ultimate client. diverse applications engineering, hardware industry. article Anurag Choudhary fault diagnosis method improved functionality (CNN) nature-inspired Artificial Bee Colony (ABCO) algorithm. diagnostic introduces analyses possible mechanical electrical faults IM. indicates infrared thermography-based anomaly outperforms vibration acoustic-based 100% classification signify potential diagnose IM reliability robustness. Subhashree Rout called Siamese-HYNAA, Siamese population-based optimizer hypercube natural aggregation (HYNAA) candidate solutions augmenting minority class. strategy against basic SMOTE SMOTE-PSOEV, measures, ROC-AUC curves, sensitivity, specificity, accuracy, characteristic stability index, balanced F1-score, informedness, markedness execution time. Siamese-HYNAA generates promising imbalanced Dongmei Jin analyse value magnetic resonance imaging Fuzzy C-means (FCM) neonatal hypoglycaemia brain injury (HBI), explore risk factors occurrence children, guidance treatment. average minimum blood glucose (1.09 ± 0.53 mmoL/L) group lower than non-brain (1.75 0.49 mmoL/L), duration (43.1 21.07 h) higher (13.79 6.81 h), p < 0.05. conclusion, MRI FCM clustering had quality. Late feeding time, low sugar long hypoglycaemic injury. final Mahmoud Ragab facilitate problems learning. binary particle swarm (PSO) firefly (FA) such it blends best each optimized way solving said problem. suggested assessed six domains publicly available UCI repository validity. similar evolutionary-based approaches prove superiority. PSO found suitable problems. To conclude, publishes out total papers. guest editors hope would benefit readers. We want express our sincere thanks editor-in-chief allowing us organize particular issue. editorial office staffs excellent, we thank them support. thankful who made possible, thoughtful contributions.

Language: Английский

---