As market saturation and competitive pressure intensify within the banking sector, the mitigation of customer churn has emerged as a critical concern. Given that the cost of acquiring new clients substantially exceeds that of retaining existing ones, the development of highly accurate churn prediction models has become imperative. In this study, a hybrid customer churn prediction model was developed by integrating Sentence Transformers with a stacking ensemble learning architecture. Customer behavioral data containing textual content was transformed into dense vector representations through the use of Sentence Transformers, thereby capturing contextual and semantic nuances. These embeddings were combined with normalized structured features. To enhance predictive performance, a stacking ensemble method was employed to integrate the outputs of multiple base models, including random forest, Gradient Boosting Tree (GBT), and Support Vector Machine (SVM). Experimental evaluation was conducted on real-world banking data, and the proposed model demonstrated superior performance relative to conventional baseline approaches, achieving notable improvements in both accuracy and the area under the curve (AUC). Furthermore, the analysis of model outputs revealed several salient predictors of customer attrition, such as anomalous transaction behavior, prolonged inactivity, and indicators of dissatisfaction with customer service. These insights are expected to inform the development of targeted intervention strategies aimed at strengthening customer retention, improving satisfaction, and fostering long-term institutional growth and stability.

Significant advancements in artificial intelligence (AI) have transformed clinical decision-making, particularly in disease detection and management. Endometriosis, a chronic and often debilitating gynecological disorder, affects a substantial proportion of reproductive-age women and is associated with pelvic pain, infertility, and a reduced quality of life. Despite its high prevalence, non-invasive and accurate diagnostic methods remain limited, frequently resulting in delayed or missed diagnoses. In this study, a novel diagnostic framework was developed by integrating deep learning (DL) with explainable artificial intelligence (XAI) to address existing limitations in the early and non-invasive detection of endometriosis. Abdominopelvic magnetic resonance imaging (MRI) data were obtained from the Crestview Radiology Center in Victoria Island, Lagos State. Preprocessing procedures, including Digital Imaging and Communications in Medicine (DICOM)-to-PNG conversion, image resizing, and intensity normalization, were applied to standardize the imaging data. A U-Net architecture enhanced with a dual attention mechanism was employed for lesion segmentation, while Gradient-weighted Class Activation Mapping (Grad-CAM) was incorporated to visualize and interpret the model’s decision-making process. Ethical considerations, including informed patient consent, fairness in algorithmic decision-making, and mitigation of data bias, were rigorously addressed throughout the model development pipeline. The proposed system demonstrated the potential to improve diagnostic accuracy, reduce diagnostic latency, and enhance clinician trust by offering transparent and interpretable predictions. Furthermore, the integration of XAI is anticipated to promote greater clinical adoption and reliability of AI-assisted diagnostic systems in gynecology. This work contributes to the advancement of non-invasive diagnostic tools and reinforces the role of interpretable DL in the broader context of precision medicine and women's health.

The selection of optimal text embedding models remains a critical challenge in semantic textual similarity (STS) tasks, particularly when performance varies substantially across datasets. In this study, the comparative effectiveness of multiple state-of-the-art embedding models was systematically evaluated using a benchmarking framework based on established machine learning techniques. A range of embedding architectures was examined across diverse STS datasets, with similarity computations performed using Euclidean distance, cosine similarity, and Manhattan distance metrics. Performance evaluation was conducted through Pearson and Spearman correlation coefficients to ensure robust and interpretable assessments. The results revealed that GIST-Embedding-v0 consistently achieved the highest average correlation scores across all datasets, indicating strong generalizability. Nevertheless, MUG-B-1.6 demonstrated superior performance on datasets 2, 6, and 7, while UAE-Large-V1 outperformed other models on datasets 3 and 5, thereby underscoring the influence of dataset-specific characteristics on embedding model efficacy. These findings highlight the importance of adopting a dataset-aware approach in embedding model selection for STS tasks, rather than relying on a single universal model. Moreover, the observed performance divergence suggests that embedding architectures may encode semantic relationships differently depending on domain-specific linguistic features. By providing a detailed evaluation of model behavior across varied datasets, this study offers a methodological foundation for embedding selection in downstream NLP applications. The implications of this research extend to the development of more reliable, scalable, and context-sensitive STS systems, where model performance can be optimized based on empirical evidence rather than heuristics. These insights are expected to inform future investigations on embedding adaptation, hybrid model integration, and meta-learning strategies for semantic similarity tasks.

Sentiment analysis in legal documents presents significant challenges due to the intricate structure, domain-specific terminology, and strong contextual dependencies inherent in legal texts. In this study, a novel hybrid framework is proposed, integrating Graph Attention Networks (GATs) with domain-specific embeddings, i.e., Legal Bidirectional Encoder Representations from Transformers (LegalBERT) and an aspect-oriented sentiment classification approach to improve both predictive accuracy and interpretability. Unlike conventional deep learning models, the proposed method explicitly captures hierarchical relationships within legal texts through GATs while leveraging LegalBERT to enhance domain-specific semantic representation. Additionally, auxiliary features, including positional information and topic relevance, were incorporated to refine sentiment predictions. A comprehensive evaluation conducted on diverse legal datasets demonstrates that the proposed model achieves state-of-the-art performance, attaining an accuracy of 93.1% and surpassing existing benchmarks by a significant margin. Model interpretability was further enhanced through SHapley Additive exPlanations (SHAP) and Legal Context Attribution Score (LCAS) techniques, which provide transparency into decision-making processes. An ablation study confirms the critical contribution of each model component, while scalability experiments validate the model’s efficiency across datasets ranging from 10,000 to 200,000 sentences. Despite increased computational demands, strong robustness and scalability are exhibited, making this framework suitable for large-scale legal applications. Future research will focus on multilingual adaptation, computational optimization, and broader applications within the field of legal analytics.

The restoration of blurred images remains a critical challenge in computational image processing, necessitating advanced methodologies capable of reconstructing fine details while mitigating structural degradation. In this study, an innovative image restoration framework was introduced, employing Complex Interval Pythagorean Fuzzy Sets (CIPFSs) integrated with mathematically structured transformations to achieve enhanced deblurring performance. The proposed methodology initiates with the geometric correction of pixel-level distortions induced by blurring. A key innovation lies in the incorporation of CIPFS-based entropy, which is synergistically combined with local statistical energy to enable robust blur estimation and adaptive correction. Unlike traditional fuzzy logic-based approaches, CIPFS facilitates a more expressive modeling of uncertainty by leveraging complex interval-valued membership functions, thereby enabling nuanced differentiation of blur intensity across image regions. A fuzzy inference mechanism was utilized to guide the refinement process, ensuring that localized corrections are adaptively applied to degraded regions while leaving undistorted areas unaffected. To preserve edge integrity, a geometric step function was applied to reinforce structural boundaries and suppress over-smoothing artifacts. In the final restoration phase, structural consistency is enforced through normalization and regularization techniques to ensure coherence with the original image context. Experimental validations demonstrate that the proposed model delivers superior image clarity, improved edge sharpness, and reduced visual artifacts compared to state-of-the-art deblurring methods. Enhanced robustness against varying blur patterns and noise intensities was also confirmed, indicating strong generalization potential. By unifying the expressive power of CIPFS with analytically driven restoration strategies, this approach contributes a significant advancement to the domain of image deblurring and restoration under uncertainty.

Stance, a critical discourse marker, reflects the expression of attitudes, feelings, evaluations, or judgments by speakers or writers toward a topic or other participants in a conversation. This study investigates the manifestation of stance in the discourse of four prominent artificial intelligence (AI) chatbots—ChatGPT, Gemini, MetaAI, and Bing Copilot—focusing on three dimensions: interpersonal stance (how chatbots perceive one another), epistemic stance (their relationship to the topic of discussion), and style stance (their communicative style). Through a systematic analysis, it is revealed that these chatbots employ various stance markers, including hedging, self-mention, power dominance, alignment, and face-saving strategies. Notably, the use of face-saving framing by AI models, despite their lack of a genuine “face,” highlights the distinction between authentic interactional intent and the reproduction of linguistic conventions. This suggests that stance in AI discourse is not a product of subjective intent but rather an inherent feature of natural language. However, this study extends the discourse by examining stance as a feature of chatbot-to-chatbot communication rather than human-AI interactions, thereby bridging the gap between human linguistic behaviors and AI tendencies. It is concluded that stance is not an extraneous feature of discourse but an integral and unavoidable aspect of language use, which chatbots inevitably replicate. In other words, if chatbots must use language, then pragmatic features like stance are inevitable. Ultimately, this raises a broader question: Is it even possible for a chatbot to produce language devoid of stance? The implications of this research underscore the intrinsic connection between language use and pragmatic features, suggesting that stance is an inescapable component of any linguistic output, including that of AI systems.

This study investigates the recognition of seven primary human emotions—contempt, anger, disgust, surprise, fear, happiness, and sadness—based on facial expressions. A transfer learning approach was employed, utilizing three pre-trained convolutional neural network (CNN) architectures: AlexNet, VGG16, and ResNet50. The system was structured to perform facial expression recognition (FER) by incorporating three key stages: face detection, feature extraction, and emotion classification using a multiclass classifier. The proposed methodology was designed to enhance pattern recognition accuracy through a carefully structured training pipeline. Furthermore, the performance of the transfer learning models was compared using a multiclass support vector machine (SVM) classifier, and extensive testing was planned on large-scale datasets to further evaluate detection accuracy. This study addresses the challenge of spontaneous FER, a critical research area in human-computer interaction, security, and healthcare. A key contribution of this study is the development of an efficient feature extraction method, which facilitates FER with minimal reliance on extensive datasets. The proposed system demonstrates notable improvements in recognition accuracy compared to traditional approaches, significantly reducing misclassification rates. It is also shown to require less computational time and resources, thereby enhancing its scalability and applicability to real-world scenarios. The approach outperforms conventional techniques, including SVMs with handcrafted features, by leveraging the robust feature extraction capabilities of transfer learning. This framework offers a scalable and reliable solution for FER tasks, with potential applications in healthcare, security, and human-computer interaction. Additionally, the system’s ability to function effectively in the absence of a caregiver provides significant assistance to individuals with disabilities in expressing their emotional needs. This research contributes to the growing body of work on facial emotion recognition and paves the way for future advancements in artificial intelligence-driven emotion detection systems.

Drought, a complex natural phenomenon with profound global impacts, including the depletion of water resources, reduced agricultural productivity, and ecological disruption, has become a critical challenge in the context of climate change. Effective drought prediction models are essential for mitigating these adverse effects. This study investigates the contribution of various data preprocessing steps—specifically class imbalance handling and dimensionality reduction techniques—to the performance of machine learning models for drought prediction. Synthetic Minority Over-sampling Technique (SMOTE) and near miss sampling methods were employed to address class imbalances within the dataset. Additionally, Principal Component Analysis (PCA) and Linear Discriminant Analysis (LDA) were applied for dimensionality reduction, aiming to improve computational efficiency while retaining essential features. Decision tree algorithms were trained on the preprocessed data to assess the impact of these preprocessing techniques on model accuracy, precision, recall, and F1-score. The results indicate that the SMOTE-based sampling approach significantly enhances the overall performance of the drought prediction model, particularly in terms of accuracy and robustness. Furthermore, the combination of SMOTE, PCA, and LDA demonstrates a substantial improvement in model reliability and generalizability. These findings underscore the critical importance of carefully selecting and applying appropriate data preprocessing techniques to address class imbalances and reduce feature space, thus optimizing the performance of machine learning models in drought prediction. This study highlights the potential of preprocessing strategies in improving the predictive capabilities of models, providing valuable insights for future research in climate-related prediction tasks.

This study introduces an advanced approach to the automated detection of skin tanning, leveraging image processing and computer vision techniques to accurately assess tanning levels. A method was proposed in which skin tone variations were analyzed by comparing a reference image with a current image of the same subject. This approach establishes a reliable framework for estimating tanning levels through a sequence of image preprocessing, skin segmentation, dominant color extraction, and tanning assessment. The hue-saturation-value (HSV) color space was employed to quantify these variations, with particular emphasis placed on the saturation component, which is identified as a critical factor for tanning detection. This novel focus on the saturation component offers a robust and objective alternative to traditional visual assessment methods. Additionally, the potential integration of machine learning techniques to enhance skin segmentation and improve image analysis accuracy was explored. The proposed framework was positioned within an Internet of Things (IoT) ecosystem for real-time monitoring of sun safety, providing a practical application for both individual and public health contexts. Experimental results demonstrate the efficacy of the proposed method in distinguishing various tanning levels, thereby offering significant advancements in the fields of cosmetic dermatology, public health, and preventive medicine. These findings suggest that the integration of image processing, computer vision, and machine learning can provide a powerful tool for the automated assessment of skin tanning, with broad implications for real-time health monitoring and the prevention of overexposure to ultraviolet (UV) radiation.

In recent years, representing computer vision data in tensor form has become an important method of data representation. However, due to the limitations of signal acquisition devices, the actual data obtained may be damaged, such as image loss, noise interference, or a combination of both. Using Low-Rank Tensor Completion (LRTC) techniques to recover missing or corrupted tensor data has become a hot research topic. In this paper, we adopt a tensor coupled total variation (t-CTV) norm based on t-SVD as the minimization criterion to capture the combined effects of low-rank and local piecewise smooth priors, thus eliminating the need for balance parameters in the process. At the same time, we utilize the Non-Local Means (NLM) denoiser to smooth the image and reduce noise by leveraging the nonlocal self-similarity of the image. Furthermore, an Alternating Direction Method of Multipliers (ADMM) algorithm is designed for the proposed optimization model, NLM-TCTV. Extensive numerical experiments on real tensor data (including color, medical, and satellite remote sensing images) show that the proposed method has good robustness, performs well in noisy images, and surpasses many existing methods in both quality and visual effects.

This study explores hybrid deep learning architectures for the classification of financial sentiment, focusing on the integration of the Convolutional Neural Network (CNN) with the Support Vector Machine (SVM) and the Random Forest (RF). CNN, with its powerful feature extraction capabilities, was combined with SVM’s ability to handle non-linear decision boundaries, while RF enhanced model generalization through ensemble learning. The proposed hybrid frameworks addressed two fundamental challenges in sentiment analysis: overfitting and class imbalance. These challenges were mitigated, resulting in improved model accuracy and reliability compared to standalone methods. Empirical evaluations demonstrated that the CNN-SVM model achieved competitive or superior validation accuracy and loss, indicating its suitability for precise financial sentiment classification. By enabling more accurate sentiment categorization, the model provides actionable insights for financial analysts and investors, thereby supporting better market assessment and investment decision-making. Future work is suggested to incorporate advanced techniques such as adversarial training and domain-specific pre-trained models to further enhance model performance.

Swarm intelligence (SI) has emerged as a transformative approach in solving complex optimization problems by drawing inspiration from collective behaviors observed in nature, particularly among social animals and insects. Ant Colony Optimization (ACO), a prominent subclass of SI algorithms, models the foraging behavior of ant colonies to address a range of challenging combinatorial problems. Originally introduced in 1992 for the Traveling Salesman Problem (TSP), ACO employs artificial pheromone trails and heuristic information to probabilistically guide solution construction. The artificial ants within ACO algorithms engage in a stochastic search process, iteratively refining solutions through the deposition and evaporation of pheromone levels based on previous search experiences. This review synthesizes the extensive body of research that has since advanced ACO from its initial ant system (AS) model to sophisticated algorithmic variants. These advances have both significantly enhanced ACO's practical performance across various application domains and contributed to a deeper theoretical understanding of its mechanics. The focus of this study is placed on the behavioral foundations of ACO, as well as on the metaheuristic frameworks that enable its versatility and robustness in handling large-scale, computationally intensive tasks. Additionally, this study highlights current limitations and potential areas for future exploration within ACO, aiming to facilitate a comprehensive understanding of this dynamic field of swarm-based optimization.

Radar warning receivers (RWRs) are critical for swiftly and accurately identifying potential threats in complex electromagnetic environments. Numerous methods have been developed over the years, with recent advances in artificial intelligence (AI) significantly enhancing RWR capabilities. This study presents a machine learning-based approach for emitter identification within RWR systems, leveraging a comprehensive radar signal library. Key parameters such as signal frequency, pulse width, pulse repetition frequency (PRF), and beam width were extracted from pulsed radar signals and utilized in various machine learning algorithms. The preprogramming phase of RWRs was optimized through the application of multiple classification algorithms, including k-Nearest Neighbors (KNN), Decision Tree (DT), the ensemble learning method, support vector machine (SVM), and Artificial Neural Network (ANN). These algorithms were compared against conventional methods to evaluate their performance. The machine learning models demonstrated a high degree of accuracy, achieving over 95% in training phases and exceeding 99% in test simulations. The findings highlight the superiority of machine learning algorithms in terms of speed and precision when compared to traditional approaches. Furthermore, the flexibility of machine learning techniques to adapt to diverse problem sets underscores their potential as a preferred solution for future RWR applications. This study suggests that the integration of machine learning into RWR emitter identification not only enhances the operational efficiency of electronic warfare (EW) systems but also represents a significant advancement in the field. The increasing relevance of machine learning in recent years positions it as a promising tool for addressing complex signal processing challenges in EW.