Integrating Machine Learning and Material Feeding Systems for Competitive Advantage in Manufacturing

Applied Sciences, Journal Year: 2025, Volume and Issue: 15(2), P. 980 - 980

Published: Jan. 20, 2025

In contemporary business environments, manufacturing companies must continuously enhance their performance to ensure competitiveness. Material feeding systems are of pivotal importance in the optimization productivity, with attendant improvements quality, reduction costs, and minimization delivery times. This study investigates selection material methods, including Kanban, line-storage, call-out, kitting systems, within a company. The research employs six machine learning (ML) algorithms—logistic regression (LR), decision trees (DT), random forest (RF), support vector machines (SVM), K-nearest neighbors (K-NN), artificial neural networks (ANN)—to develop multi-class classification model for system selection. Utilizing dataset comprising 2221 materials an 8-fold cross-validation technique, ANN exhibits superior across all evaluation metrics. Shapley values analysis is employed elucidate influence input parameters process systems. provides comprehensive framework selection, integrating advanced ML models practical insights. makes significant contribution field by enhancing decision-making processes, optimizing resource utilization, establishing foundation future studies on adaptive scalable strategies dynamic industrial environments.

Language: Английский

---

Evaluating Sparse Feature Selection Methods: A Theoretical and Empirical Perspective

Applied Sciences, Journal Year: 2025, Volume and Issue: 15(7), P. 3752 - 3752

Published: March 29, 2025

This paper analyzes two main categories of feature selection: filter methods (such as minimum redundancy maximum relevance, CHI2, Kruskal–Wallis, and ANOVA) embedded alternating direction method multipliers (BP_ADMM), least absolute shrinkage selection operator, orthogonal matching pursuit). The mathematical foundations inspired by compressed detection are presented, highlighting how the principles sparse signal recovery can be applied to identify most relevant features. results have been obtained using biomedical databases. used algorithms have, their starting point, notion sparsity, but version implemented tested in this work is adapted for selection. experimental show that BP_ADMM achieves highest classification accuracy (77% arrhythmia_database 100% oncological_database), surpassing both full set other study, which makes it optimal option. analysis shows strike a balance between efficiency selecting features during model training, unlike filtering methods, ignore interactions. Although more accurate, slower depend on chosen algorithm. less comprehensive than wrapper they offer strong trade-off speed performance when computational resources allow it.

Language: Английский

---

Strategies to Improve the Robustness and Generalizability of Deep Learning Segmentation and Classification in Neuroimaging

BioMedInformatics, Journal Year: 2025, Volume and Issue: 5(2), P. 20 - 20

Published: April 14, 2025

Artificial Intelligence (AI) and deep learning models have revolutionized diagnosis, prognostication, treatment planning by extracting complex patterns from medical images, enabling more accurate, personalized, timely clinical decisions. Despite its promise, challenges such as image heterogeneity across different centers, variability in acquisition protocols scanners, sensitivity to artifacts hinder the reliability integration of models. Addressing these issues is critical for ensuring accurate practical AI-powered neuroimaging applications. We reviewed summarized strategies improving robustness generalizability segmentation classification neuroimages. This review follows a structured protocol, comprehensively searching Google Scholar, PubMed, Scopus studies on neuroimaging, task-specific applications, model attributes. Peer-reviewed, English-language brain imaging were included. The extracted data analyzed evaluate implementation effectiveness techniques. study identifies key enhance including regularization, augmentation, transfer learning, uncertainty estimation. These approaches address major domain shifts, consistent performance diverse settings. technical this can improve their real-world practice.

Language: Английский

---

Using the β/α Ratio to Enhance Odor-Induced EEG Emotion Recognition

Applied Sciences, Journal Year: 2025, Volume and Issue: 15(9), P. 4980 - 4980

Published: April 30, 2025

Emotion recognition using an odor-induced electroencephalogram (EEG) has broad applications in human-computer interaction. However, existing studies often rely on subjective self-reporting to label emotion, lacking objective verification. While the β/α ratio been identified as a potential indicator of arousal EEG spectral analysis, its value emotion remains underexplored. This study ensured authenticity emotions through and analysis 50 adults after inhaling sandalwood essential oil (SEO) or bergamot (BEO). Classification models were built discriminant (DA), support vector machine (SVM), random forest (RF) algorithms identify low high emotions. Notably, this introduced novel frequency domain feature enhance model performance for first time. Both indicated that SEO promotes relaxation, whereas BEO enhances attentiveness. In testing, incorporating enhanced all models, with accuracy DA, SVM, RF increasing from 70%, 75%, 85% 80%, 95%, respectively. validated by employing combination methods highlighted importance along dimension.

Language: Английский

---

A Resource-Efficient Multi-Entropy Fusion Method and Its Application for EEG-Based Emotion Recognition

Entropy, Journal Year: 2025, Volume and Issue: 27(1), P. 96 - 96

Published: Jan. 20, 2025

Emotion recognition is an advanced technology for understanding human behavior and psychological states, with extensive applications mental health monitoring, human–computer interaction, affective computing. Based on electroencephalography (EEG), the biomedical signals naturally generated by brain, this work proposes a resource-efficient multi-entropy fusion method classifying emotional states. First, Discrete Wavelet Transform (DWT) applied to extract five brain rhythms, i.e., delta, theta, alpha, beta, gamma, from EEG signals, followed acquisition of features, including Spectral Entropy (PSDE), Singular Spectrum (SSE), Sample (SE), Fuzzy (FE), Approximation (AE), Permutation (PE). Then, such entropies are fused into matrix represent complex dynamic characteristics EEG, denoted as Brain Rhythm Matrix (BREM). Next, Dynamic Time Warping (DTW), Mutual Information (MI), Spearman Correlation Coefficient (SCC), Jaccard Similarity (JSC) measure similarity between unknown testing BREM data positive/negative samples classification. Experiments were conducted using DEAP dataset, aiming find suitable scheme regarding measures, time windows, input numbers channel data. The results reveal that DTW yields best performance in measures 5 s window. In addition, single-channel mode outperforms single-region mode. proposed achieves 84.62% 82.48% accuracy arousal valence classification tasks, respectively, indicating its effectiveness reducing dimensionality computational complexity while maintaining over 80%. Such performances remarkable when considering limited resources concern, which opens possibilities innovative entropy can help design portable EEG-based emotion-aware devices daily usage.

Language: Английский

---