Deep learning-based human activity recognition using CNN, ConvLSTM, and LRCN

International Journal of Cognitive Computing in Engineering, Journal Year: 2024, Volume and Issue: 5, P. 259 - 268

Published: Jan. 1, 2024

Human activity recognition (HAR) plays a crucial role in assisting the elderly and individuals with vascular dementia by providing support monitoring for their daily activities. This paper presents deep learning (DL)-based approach to HAR, leveraging convolutional neural network (CNN), long short-term memory (ConvLSTM), long-term recurrent (LRCN) architectures. These models are designed extract spatial features capture temporal dependencies video data, enhancing accuracy of classification. We conducted experiments on UCF50 HMDB51 datasets, encompassing diverse human Our evaluation demonstrates that ConvLSTM model achieves an 82% 68% HMDB51, while LRCN gives accuracies 93.44% 71.55%, respectively. Finally, CNN outperforms rate 99.58% 92.70% datasets. significant improvements showcase effectiveness integrating networks HAR tasks. research contributes advancing systems potential applications healthcare, assisted living, surveillance. By accurately recognizing activities, our can assist remote patient monitoring, fall detection, public safety initiatives. findings underscore importance DL quality life various contexts.

Language: Английский

---

Toward reliable diabetes prediction: Innovations in data engineering and machine learning applications

Digital Health, Journal Year: 2024, Volume and Issue: 10

Published: Jan. 1, 2024

Objective Diabetes is a metabolic disorder that causes the risk of stroke, heart disease, kidney failure, and other long-term complications because diabetes generates excess sugar in blood. Machine learning (ML) models can aid diagnosing at primary stage. So, we need an efficient ML model to diagnose accurately. Methods In this paper, effective data preprocessing pipeline has been implemented process random oversampling balance data, handling imbalance distributions observational more sophisticatedly. We used four different datasets conduct our experiments. Several algorithms were determine best predict faultlessly. Results The performance analysis demonstrates among all algorithms, forest surpasses current works with accuracy rate 86% 98.48% for Dataset 1 2; extreme gradient boosting decision tree surpass 99.27% 100% 3 4, respectively. Our proposal increase by 12.15% compared without preprocessing. Conclusions This excellent research finding indicates proposed might be employed produce accurate predictions supplement preventative interventions reduce incidence its associated costs.

Language: Английский

---

ResInceptNet-SA: A Network Traffic Intrusion Detection Model Fusing Feature Selection and Balanced Datasets

Applied Sciences, Journal Year: 2025, Volume and Issue: 15(2), P. 956 - 956

Published: Jan. 19, 2025

Network intrusion detection models are vital techniques for ensuring cybersecurity. However, existing face several challenges, such as insufficient feature extraction capabilities, dataset imbalance, and suboptimal accuracy. In this paper, a new type of model (ResIncepNet-SA) based on InceptionNet, Resnet, convolutional neural networks with self-attention mechanism was proposed to detect network intrusions. The used the PCA-ADASYN algorithm compress traffic features, extract high-correlation datasets, oversample balance datasets classify abnormal traffic. experimental results show that accuracy, precision, recall, F1-score ResIncepNet-SA using NSL-KDD reach 0.99366, 0.99343, 0.99339, 0.99338, respectively. This enhances accuracy outperforms when applied imbalanced offering solution detection.

Language: Английский

---

A hybrid machine learning model for intrusion detection in wireless sensor networks leveraging data balancing and dimensionality reduction

Scientific Reports, Journal Year: 2025, Volume and Issue: 15(1)

Published: Feb. 7, 2025

Intrusion detection systems are essential for securing wireless sensor networks (WSNs) and Internet of Things (IoT) environments against various threats. This study presents a novel hybrid machine learning (ML) model that integrates KMeans-SMOTE (KMS) data balancing principal component analysis (PCA) dimensionality reduction, evaluated using the WSN-DS TON-IoT datasets. The employs classifiers such as Decision Tree Classifier, Random Forest Classifier (RFC), gradient boosting techniques like XGBoost (XGBC) to enhance accuracy efficiency. proposed (KMS + PCA RFC) approach achieves remarkable performance, with an 99.94% f1-score on dataset. For dataset, it 99.97% 99.97%, outperforming traditional SMOTE TomekLink Generative Adversarial Network-based techniques. addresses class imbalance high-dimensionality challenges, providing scalable robust intrusion detection. Complexity reveals reduces training prediction times, making suitable real-time applications.

Language: Английский

---

Enhancing malware detection with feature selection and scaling techniques using machine learning models

Scientific Reports, Journal Year: 2025, Volume and Issue: 15(1)

Published: March 17, 2025

Abstract The increasing prevalence of malware presents a critical challenge to cybersecurity, emphasizing the need for robust detection methods. This study uses binary tabular classification dataset evaluate impact feature selection, scaling, and machine learning (ML) models on detection. methodology involves experimenting with three scaling techniques (no normalization, min-max scaling), selection methods Linear Discriminant Analysis (LDA), Principal Component (PCA)), twelve ML models, including traditional algorithms ensemble A publicly available 11,598 samples 139 features is utilized, model performance assessed using metrics such as accuracy, precision, recall, F1-score, AUC-ROC. Results reveal that Light Gradient Boosting Machine (LGBM) achieves highest accuracy 97.16% when PCA either or normalization are applied. Additionally, consistently outperform demonstrating their effectiveness in enhancing These findings offer valuable insights into optimizing preprocessing strategies developing reliable efficient systems.

Language: Английский

---

An integrated multistage ensemble machine learning model for fraudulent transaction detection

Journal Of Big Data, Journal Year: 2024, Volume and Issue: 11(1)

Published: Nov. 22, 2024

Abstract Fraudulent transactions continue to pose a concern for financial institutions and organizations, necessitating the development of effective detection tools. Identification prevention fraudulent depend heavily on credit card fraud. Even though instances fraud are uncommon, they can nonetheless cause significant losses because high cost transactions. When is discovered early on, investigators act quickly stop additional losses. But investigation process takes while, there only so many warnings that be looked through in detail given day. Thus, model’s main goal minimize false alarms missed situations while producing accurate alerts. To improve identification, we provide this study an integrated multistage ensemble Machine Learning (IMEML) model incorporates various models intelligently, such as Ensemble Independent Classifier (EIC), Bagging (EBC), ML (EMC). In order overcome problem data imbalance, use number methods-including Instant Hardness Threshold with EMC (IHT+EMC), Cluster Centroids (CC), Randon Under Sampler (RUS)-that go beyond traditional methods. We run our studies 284,807-transaction dataset made available public. The accuracy rates 99.94%, 99.91%, 99.14%, 99.52%, perfect 100% accuracy, precision, recall, f1-score, AUC score, respectively, achieved by suggested model, demonstrating remarkable performance scores. For real-world applications, EIBMC sets new benchmark identifying high-frequency scenarios outperforming cutting-edge techniques.

Language: Английский

---

A hybrid cardiovascular arrhythmia disease detection using ConvNeXt-X models on electrocardiogram signals

Scientific Reports, Journal Year: 2024, Volume and Issue: 14(1)

Published: Dec. 5, 2024

Cardiovascular arrhythmia, characterized by irregular heart rhythms, poses significant health risks, including stroke and failure, making accurate early detection critical for effective treatment. Traditional methods often struggle with challenges such as imbalanced datasets, limiting their ability to identify rare arrhythmia types. This study proposes a novel hybrid approach that integrates ConvNeXt-X deep learning models advanced data balancing techniques improve classification accuracy. Specifically, we evaluated three ConvNeXt variants—ConvNeXtTiny, ConvNeXtBase, ConvNeXtSmall—combined Random Oversampling (RO) SMOTE-TomekLink (STL) on the MIT-BIH Arrhythmia Database. Experimental results demonstrate ConvNeXtTiny model paired STL achieved highest accuracy of 99.75%, followed RO at 99.72%. The technique consistently enhanced minority class overall performance across models, ConvNeXtBase ConvNeXtSmall achieving accuracies 99.69% 99.72%, respectively. These findings highlight efficacy when coupled robust techniques, in reliable precise detection. methodology holds potential improving diagnostic supporting clinical decision-making healthcare.

Language: Английский

---

A stacked ensemble approach to detect cyber attacks based on feature selection techniques

International Journal of Cognitive Computing in Engineering, Journal Year: 2024, Volume and Issue: 5, P. 316 - 331

Published: Jan. 1, 2024

The exponential growth of data and increased reliance on interconnected systems have heightened the need for robust network security. Cyber-Attack Detection Systems (CADS) are essential identifying mitigating threats through traffic analysis. However, effectiveness CADS is highly dependent selecting pertinent features. This research evaluates impact three feature selection techniques—Recursive Feature Elimination (RFE), Mutual Information (MI), Lasso Selection (LFS)—on performance. We propose a novel stacked ensemble classification approach, combining Random Forest, XGBoost, Extra-Trees classifiers with Logistic Regression meta-model. Performance assessed using CICIDS2017 NSL-KDD datasets. Results show that RFE achieves 100% accuracy Brute Force attacks, 99.99% Infiltration Web Attacks CICIDS2017, 99.95% all attacks NSL-KDD, marking significant improvement over traditional methods. study demonstrates optimizing leveraging diverse can substantially enhance CADS, providing stronger protection against evolving cyber threats.

Language: Английский

---

usfAD based effective unknown attack detection focused IDS framework

Scientific Reports, Journal Year: 2024, Volume and Issue: 14(1)

Published: Nov. 24, 2024

Language: Английский

---

A dual-tier adaptive one-class classification IDS for emerging cyberthreats

Computer Communications, Journal Year: 2024, Volume and Issue: 229, P. 108006 - 108006

Published: Nov. 14, 2024

Language: Английский

---