Artificial intelligence (AI) is increasingly associated with changes in e-commerce development, including demand generation, operational processes, and consumer interaction. The aim of the article is to examine and quantify the relationship between AI adoption and e-commerce development in 2023–2025 in an international context, with an emphasis on Ukraine’s position among European markets. The methodological framework of the study is a quantitative comparative analysis of data for Germany, France, Italy, Spain, Poland, the Netherlands, and Ukraine. An integral E-Commerce Index and an AI Adoption Index were created based on statistical sources, which were normalized in the range from 0 to 1. The hypothesis was tested by using correlation and regression analysis, as well as clustering of countries by a combination of e-commerce and AI maturity levels. Given the correlational design of the study, the findings are interpreted as evidence of statistical association rather than causation. The results show a persistent positive and statistically significant relationship between the AI adoption and the development of e-commerce levels. The leaders in terms of the AI Adoption Index are the Netherlands (0.88) and Germany (0.85), while Ukraine demonstrates the earliest phase of integration (0.25). The findings suggest that AI adoption is a significant correlate of e-commerce development rather than merely an auxiliary technological feature. The practical value of the results is the possibility of their use for building state policy of digital transformation and business strategies in the context of international competition.

This study develops a Python-based hybrid forecasting framework that integrates classical Decline Curve Analysis (DCA) with Support Vector Regression (SVR) to improve the reliability of oil production forecasting and the estimation of Estimated Remaining Reserves (ERR), which are subsequently used to derive cumulative CO₂ emission potential. Historical production data are segmented to represent boundary-dominated flow conditions, enhancing the stability and physical consistency of decline parameters. SVR is then applied to correct localized deviations in the historical response that are not captured by conventional DCA formulations, yielding improved decline representations for forecasting. The hybrid DCA–SVR model achieves a more coherent historical match and reduced average residual errors compared to standalone exponential, hyperbolic, and harmonic decline models. Using the hybrid framework, ERR estimates of 4.92×10⁵, 1.74×10⁶, and 3.62×10⁶ STB were obtained for the exponential, hyperbolic, and harmonic decline scenarios, respectively, corresponding to cumulative CO₂ emission potentials of approximately 213,036, 753,420, and 1,567,460 tons of CO₂. These scenario-based results provide a bounded range of emission outcomes, explicitly demonstrating that uncertainty in decline model selection propagates directly into long-term ERR and CO₂ emission estimates. Overall, the proposed hybrid machine learning–decline curve analysis (ML-DCA) framework offers a transparent and reproducible approach for linking production forecasting with quantitative CO₂ impact assessment in data-driven reservoir management.

With the growing digitization of financial services, fraudulent activities in online transactions are becoming increasingly sophisticated and widespread. An automated and intelligent financial transaction fraud detection system is necessary to properly detect and categorize fraudulent incidents in order to address this challenge. The majority of previous research has been on using binary classification to identify fraudulent transactions. Previous efforts were often hindered by class imbalance, inadequate feature selection, suboptimal prediction accuracy, and insufficient hyperparameter tuning. This paper suggests a hybrid deep learning framework for multi-class financial fraud detection that combines Convolutional Neural Network (CNN) and Closed-form Continuous-time (CfC) models in order to handle these problems. By analyzing transactional and behavioral characteristics, the model is intended to categorize fraud situations into distinct groups, including Phishing, Account Takeover (ATO), Card Skimming, and No Fraud. The proposed CfC-CNN model was not only trained but also tested on a large transaction augmented dataset that was enhanced with contextual characteristics. This work uses the IBM financial fraud dataset and considers 5 million transactions. This work investigated four classes such as phishing, ATO, card skimming, and no fraud. Label encoding was used as a preprocessing method for the contextual and category information. The dataset was refined for best performance by feature selection using Chi-Square and the random search method for hyper parameter tweaking. The proposed CfC-CNN model achieved at least .03 percent accuracy gain and .49 percent F1 score gain over the compared previous works.

Digital technologies optimize certain aspects of customs operations, but also change the functional purpose of customs, raising its role in ensuring the transparency of international trade. The aim of the study was to assess the impact of factors measured by the Organisation for Economic Co-operation and Development (OECD) trade facilitation indicators on the transparency of customs operations. The study employed correlation, regression analyses, and analysis of variance (ANOVA). The study provides an integrated empirical analysis of the complex impact of technological, operational, and regulatory factors on the transparency of customs operations. The results revealed that the automation of customs procedures with a regression coefficient for the corresponding indicator of 0.3132 had the strongest positive impact on the availability of information (transparency of customs operations). A strong impact of the opportunities and methods for appealing administrative decisions of border authorities (0.2688), as well as simplifying customs control procedures and introducing “single windows” (0.2384) was observed. A less considerable, but statistically significant impact was shown by compliance with the discipline of fees, payments, and fines (0.1438) and simplification and standardization of trade documentation (0.1366). It is concluded that strategies for increasing the transparency of customs operations should focus on the above factors, while increasing the effectiveness of areas that did not show a significant impact on transparency. These are the involvement of the trade community, interaction between internal and external border agencies, the development of Advance rulings, as well as ensuring good governance and impartiality. The effectiveness of these areas should be increased by improving the mechanisms for their provision at the national and international levels. The results of the study are useful for determining priority areas of strategies for increasing the transparency of customs operations, especially regarding the development of digital platforms, for example, accelerating automation, developing single windows, etc.

Keylogging has become a crucial attacking strategy adopted by cyber criminals to fetch keystrokes of legitimate users during their interaction with computer systems. The data gathered serves as a powerful weapon for attackers to exploit. Existing techniques for detecting or preventing keylogging operations rely on mundane systems or traditional machine learning algorithms that lack deep intelligence to identify and mitigate activities of keyloggers accurately due to the volume of data that is generated from keystrokes with simple interactions from the users. This study developed a deep learning-based predictive model capable of preventing keylogging attacks in human-computer interaction using hyperband hyperparameter tuning and a hybrid genetic algorithm on stochastic gradient descent with Adam-based LSTM optimization on a UNSW-NB keylogging dataset with nine families of attacks to build the deep learning model called HH-GASA-OLSTM. From the experiments, the proposed predictive model on an 80:20 validation ratio gives the best result on six performance metrics, which are loss function of 0.0187, accuracy of 99.45%, precision of 0.9930, recall of 0.9945, F1-Score of 0.9938, MCC of 0.9889, RMSE of 0.0054, MAE of 0.0054 and MSE of 0.00003. These results are suitable for the accurate detection and prevention of keylogging attacks in HCI. The study therefore recommends further studies; the implementation of other hyperparameter tuning approaches with other parameter optimization techniques using other deep learning architectures, such as radial basis function network and deep belief network, for improved model optimization.

In an increasingly competitive world, retail companies are implementing marketing strategies aimed at gaining customer preference by building lasting emotional relationships. This facilitates the conversion of so-called followers into loyal customers and brand ambassadors, thereby increasing return on investment (ROI), facilitating direct feedback to improve the product mix, and reducing customer acquisition costs over time, generating competitive advantages. The purpose of this systematic review is to analyze recent scientific production (2020-2025) on the application of Artificial Intelligence (AI) for the personalization of engagement in B2C business models, in order to analyze the predominant techniques, evaluate their impact on the consumer and determine the critical factors that condition long-term customer loyalty; the methodology used consisted of a systematic review in the main scientific databases in publications of high-impact journals. The main findings suggest that personalization tends to be dominated by machine learning models focused on algorithmic efficiency; however, a critical asymmetry was observed: the literature shows a hyper-focus on quantitative studies and functional efficiency, while studies on trust and ethics are the least relevant. Additionally, knowledge is biased by Asian dominance (53%) and the "novelty effect," limiting the applicability of strategies in markets with high demands for transparency, such as the American market. Furthermore, an emerging trend toward an Ethical-Relational Model has been observed in the literature. It is concluded that moving beyond the functional efficiency-centric approach and integrating Ethical AI as a strategic consideration may be key to fostering sustainable engagement, as evidenced by the reviewed literature.

Information geometry is a contemporary geometric exploration of the universe of information. Presently, informational theories have primarily been researched using algebraic, logical, analytical, and probabilistic approaches. Geometry, which analyses reciprocal relationships between components such as distance and curvature, could give major advantages to the science of information. In this study, we proposed a novel model for EEG signals classification based on a Riemannian space fuzzy neural network. The sample covariance matrix is utilized as an EEG signal characteristic in this suggested approach, and Riemannian mean and distance are used to categorize the matrix, therefore achieving the usage of manifold learning. The experimental findings indicate that this procedure is more successful and precise than earlier methods for recognizing messages from brain waves.

The study was determined by the growing need to quantify behavioural and institutional transformations driven by algorithmic digital finance platforms. This justified developing an integrative analytical framework for behavioural institutional assessment within artificial intelligence (AI)-driven financial ecosystems.
The aim of the study was to develop and empirically substantiate a multidimensional framework for assessing the transformative influence of algorithmic digital finance systems on financial behaviour and responsibility, integrating cognitive, institutional, and ethical dimensions of digital financial governance.
Research methods: Comparative Qualitative Critical Assessment, ITI Modelling of Behavioural and Institutional Transformation, Structural Decomposition of Robo-Advisory Systems, Optimized Logic-Structural Architecture, ITI Verification of the Optimized Framework.
The integrated research validated the systemic transformation capacity of algorithmic digital finance platforms through behavioural and institutional dimensions. Comparative analysis revealed asymmetric configurations of behavioural rationalization and responsibility formalization, emphasizing architecture-specific functional differentiation. ITI modelling quantified these interdependencies, producing average indices of behavioural transformation (0.81) and responsibility institutionalization (0.77). Structural and optimization modelling of Robo-Advisory Systems demonstrated enhanced algorithmic transparency, cognitive traceability, and compliance integrity. Final ITI verification confirmed the optimized Robo-Advisory System as the most effective configuration (ITI* = 0.86; Bₚ = 0.85; Rₚ = 0.90), substantiating its role as a scalable paradigm for ethical and cognitively rational digital financial governance.
The academic novelty of the study lies in the development of an integrated behavioural-institutional assessment framework for evaluating algorithmic digital finance systems. The Integral Transformative Index (ITI*) is not positioned as a new mathematical operator; rather, it operationalizes a theoretically grounded construct that combines behavioural transformation and responsibility institutionalization within a single analytical architecture. The innovation is determined by the original conceptual linkage, indicator taxonomy, and platform-specific interpretation of cognitive rationalization, algorithmic transparency, and institutional accountability. The Optimized Robo-Advisory System (ORAS) model embodied this construct through explainability, bias-correction, traceability, and ethical governance mechanisms.
The next stage of research will focus on the algorithmic improvement and structural enhancement of the ORAS framework developed in this study. Controlled pilot implementation will be conducted to test behavioural adaptability, ethical compliance, and regulatory resilience. Subsequent multi-phase empirical validation will ensure the full verification of cognitive, normative, and operational dimensions of the optimized model.

In autonomous driving, Outdoor Scene (OS) understanding in unfavorable weather conditions is pivotal for enabling safe navigation. Conventional studies failed to capture the intricate pairwise object relationship within scenes in OS classification. To mitigate these challenges, this research introduces a dual framework bringing together a Panoptic Cycle-Generative Adversarial Network (PC-GAN) with a Transfer Learning-based Squeeze-and-Zero-Knowledge Spatial Excitation DenseNet (TL-SZKSE-DenseNet). The suggested PC-GAN achieves realistic cross-weather image translation with semantically aligned representations in clear and degraded scenes. The TL-SZKSE-DenseNet improves feature learning with the addition of a zero-knowledge spatial excitation mechanism that selectively enhances weather-independent and context-dependent features for enhanced discrimination in crowded scenes. This generative–discriminative hybrid method effectively overcomes structured and unstructured driving domain gaps, as demonstrated on benchmark datasets such as BDD and IndiaScene365. Primarily, the OS images are generalized to adverse conditions using Panoptic Cycle-Generative Adversarial Network (PC-GAN). After that, using saliency mapping, the most significant regions in the generalized output are highlighted. In the meantime, the significant features are extracted utilizing the Orangutan Optimization Algorithm (OOA). In addition, from the detected objects, the Object Frequency and Relationship (OFR) is analyzed. Later, using TL-SZKSE-DenseNet, the OS are classified based on the OFR and the selected features. The performance of our proposed approach was tested on our newly developed IndiaScene365 dataset as well as various benchmark datasets such as BDD100k. Our newly developed dataset, IndiaScene365, includes 3000 images that entail three different weather conditions, which are foggy weather, rainy weather, and poor light conditions along with clear day weather. The proposed approach records 94.87% accuracy, 93.24%, and 94.18% sensitivity on our dataset.

Proactive prediction based channel allocation for Cognitive Radio Networks (CRNs) remains a challenging area of research. In Internet of Things (IoT) scenarios, prediction algorithms must balance resource efficiency with high accuracy. The proposed algorithm, ProCLAMUS, a Matrix Completion (MC) Based Proactive Spectrum Allocation Protocol for CRNs, couples nuclear norm matrix completion with short horizon sliding window prediction. By reconstructing sparse spectrum sensing data and down weighting inconsistent reports, ProCLAMUS enables infrequent sensing and decentralized decisions without a fusion center, reducing energy and attack surface. In a network of 40 Secondary Users (SUs), 400 Primary Users (PUs) and 400 channels with malicious data ranging from 5 to 50%, ProCLAMUS sustains the highest channel utilization (avg 89.96%) with the lowest backoff rate (3.81 s^(-1)) and sensing delay (0.58 channels/success). ProCLAMUS achieves the lowest radio energy (16.82×10^(-3) J/s), a 44–53% reduction when compared with recently proposed techniques, while using 33–43% less memory. These gains arise from sparse sensing, fusion (majority voting) and conservative allocation. The results demonstrate superior energy efficiency and robust prediction under malicious data conditions during the Spectrum Sensing Data Falsification Attack.

IoT healthcare devices often lack resources, and encrypting patient data flows securely is an active area of research. Current cryptographic solutions are typically too computationally and energy intensive to be used on constrained IoT devices. In this work we introduce HealthCrypt: an authenticated encryption scheme for smart IoT deployments in healthcare on highly-constrained microcontrollers. It consists of a novel compact SP-network paired with native authenticated encryption, removing the need to have a separate MAC pass and minimizing total cycle cost. We implement HealthCrypt on an ARM Cortex-M0 running at 48 MHz with the FELICS benchmark suite. Results show HealthCrypt uses only 90 bytes of RAM, only requires 1,620 gate equivalents (under the proposed IoT limit of 2,000 GE) of hardware area, and only uses 5.93 nJ of energy per block encryption. Measurements of its statistical security show it has an almost perfect avalanche of 0.5026, Shannon entropy only 0.0094 bits/byte away from the ideal, and good key sensitivity of 0.4949. Against competitors like PRESENT-80, SPECK-64/128, and SIMON-64/128, HealthCrypt has been rated to have the best overall IoT Healthcare Suitability Score of 79.23.

The aim of the study is to assess the impact of digitalisation, institutional transparency, and tax burden on the scale of the shadow economy in European Union (EU) countries in order to formulate strategic guidelines for Ukraine. The study employed correlation analysis, multiple regression analysis, robustness testing, and logical structure analysis. The results revealed that an increase in the Corruption Perceptions Index (CPI) by one point reduces the share of the shadow economy by 0.2604 percentage points (p = 0.0026), highlighting the critical role of institutional transparency and anti-corruption policy in the de-shadowing process. Tax burden demonstrated a positive and statistically significant relationship with the shadow economy (coefficient = 0.1843; p = 0.0125), indicating that excessive fiscal pressure may stimulate informal economic activity. The E-Government Development Index (EGDI) retained a positive but only marginally significant effect (coefficient = 50.3170; p = 0.0667), suggesting that digitalisation becomes effective primarily when embedded in public administration systems. In contrast, the overall level of digital development measured by DESI was not statistically significant after controlling for institutional and fiscal factors (coefficient = −0.2476; p = 0.1037). The constructed multifactor model demonstrated high explanatory power (R² = 0.8503), while robustness analysis confirmed the stability of the core findings across alternative model specifications. The results indicate that digitalisation alone is insufficient to reduce informality and that its effectiveness depends on institutional quality and balanced fiscal policy. The practical value of the study lies in identifying policy directions for Ukraine, including the expansion of e-government tools, digital tax administration, integrated data analytics, and anti-corruption measures aimed at reducing the shadow economy.