This paper presents a comprehensive review of handover management and load balancing optimization (LBO) in ultra-dense 5G and emerging 6G cellular networks. With the increasing deployment of small cells and the rapid growth of data traffic, these networks face significant challenges in ensuring seamless mobility and efficient resource allocation. Traditional handover and load balancing techniques, primarily designed for 4G systems, are no longer sufficient to address the complexity of heterogeneous network environments that incorporate millimeter-wave communication, Internet of Things (IoT) devices, and unmanned aerial vehicles (UAVs). The review focuses on how recent advances in artificial intelligence (AI), particularly machine learning (ML) and deep learning (DL), are being applied to improve predictive handover decisions and enable real-time, adaptive load distribution. AI-driven solutions can significantly reduce handover failures, latency, and network congestion, while improving overall user experience and quality of service (QoS). This paper surveys state-of-the-art research on these techniques, categorizing them according to their application domains and evaluating their performance benefits and limitations. Furthermore, the paper discusses the integration of intelligent handover and load balancing methods in smart city scenarios, where ultra-dense networks must support diverse services with high reliability and low latency. Key research gaps are also identified, including the need for standardized datasets, energy-efficient AI models, and context-aware mobility strategies. Overall, this review aims to guide future research and development in designing robust, AI-assisted mobility and resource management frameworks for next-generation wireless systems.

This paper formalizes the model of choosing a strategy for reducing air pollution in an urban environment. The model involves determining the optimal location of biotechnological systems – biotechnological filter systems or smart air purification devices based on solving the problem of discrete optimization, taking into consideration the forecast of the air quality index. Two subtasks have been formalized, which make it possible to form a strategy for reducing air pollution. To solve one of the subtasks, a combined selective model for predicting the time series of the Air Quality Index (CSM) was built. The combined model software suite consists of the EMD-ESM hybrid model (Empirical Mode Decomposition-Exponential Smoothing Model), the HWM additive model (Holt-Winters Model), and the adaptive TLM (Trigg-Lich Model). To verify the proposed combined selective model, the time series of air quality indices (AQI) for the city of Nur-Sultan (data from 2010‒2021, period 6 hours) were selected. As a result of verification, it was established that in the case of short-term forecasting of the air quality index time series, the EMD-ESM model has an advantage according to the criterion of a minimum root mean square error (RMSE), δ=0.11. For the case of medium-term forecasting of 3<τ≤5, the combined selective model (CSM) has the advantage. The results reported here are input data for the task of choosing strategies for reducing the volume of air pollution in the urban environment. The study’s results make it possible to increase the flexibility of the formation of strategies for reducing air pollution since they avoid restrictions on the location of cleaners in specific urban areas. The consequence is the improvement of the environmental situation in the city and the development of the region in general

This study considers the creation of a network of moss-based biotechnological purification filters under the Smart City concept. The extent of the absorption of heavy metals and gases by Sphagnopsida moss under different conditions was investigated. The efficiency of air purification with biotechnological filters was also investigated in the city of Almaty, Republic of Kazakhstan, where an excess of the permissible concentration of harmful substances in the air, according to the WHO air quality guidelines, is recorded throughout the year. Data on the level of pollution recorded with sensors located in the largest Kazakhstani cities from 21 June 2020 to 4 June 2023 were selected as the basis for calculating purification efficiency. In total, there are 220 in 73 settlements of the Republic of Kazakhstan, with 80 such sensors located in the city of Almaty. Since creating a single biotechnological filter is expensive, our task was to calculate the air purification effect in the case of increasing the number of filters placed in polluted areas. We show that 10 filters provide an air purification efficiency of 0.77%, with 100 filters providing an air purification efficiency of 5.72% and 500 filters providing an air purification efficiency of 23.11%. A biotechnological filter for air purification based on moss was designed at Astana IT University by taking into consideration the climatic features, distribution, and types of pollution in the Republic of Kazakhstan. The obtained results are essential for ensuring compliance with the standard for environmental comfort in the Republic of Kazakhstan. Additionally, the research findings and the experience of implementing a moss-based biotechnological filter can be applied to designing similar air purification systems in other cities. This is of great importance for the advancement of the field of urban science.

This study aims to investigate state-of-the-art methods for scalable database management, focusing on cost-effectiveness. The study compares two commonly used NoSQL databases, MongoDB, and Cassandra DB, with a traditional relational database management system, PostgreSQL. The comparison is made through performance evaluations conducted on multiple clusters encompassing cloud and on-premises environments. The evaluations include database operations such as insertion, selection, and deletion. The study results indicate that each database management system is optimized for particular use cases. The choice must be based on a comprehensive assessment of multiple factors, not limited to scalability and performance. The novelty of this study lies in its scalability comparison of three popular database management systems and its emphasis on cost-effectiveness while evaluating performance in different environments.

Machine Learning (ML) and Deep Learning (DL) models are increasingly deployed in critical domains such as computer vision, natural language processing (NLP), automatic speech recognition (ASR), time-series analysis, and cybersecurity. Despite their success, these models remain vulnerable to adversarial attacks that can mislead predictions and undermine system reliability. To address this challenge, we conduct a Systematic Literature Review (SLR) of adversarial attacks and defenses in ML/DL, following PRISMA guidelines. Our review analyzes 132 peer-reviewed studies published between 2017 and 2025, identified from leading digital libraries including IEEE, ACM, ScienceDirect, and SpringerLink. The objectives of this study are to: (i) identify state-of-the-art adversarial attack and defense techniques and assess their strengths and limitations; (ii) evaluate how novel threats and countermeasures are addressed across domains; and (iii) propose a unified taxonomy that integrates adversarial attacks, defenses, evaluation metrics, and application domains. Our findings show that adversarial training and certified robustness dominate defense research, but both face trade-offs in scalability and generalizability. Computer vision remains the most studied domain, while NLP, ASR, and time-series research is growing but lacks standardized benchmarks. This review contributes a taxonomy and cross-domain synthesis of adversarial ML research, highlights research gaps, and outlines future directions to advance robustness and support the development of trustworthy AI systems.

The proliferation of digital public service platforms and the expansion of e-government initiatives have significantly increased the volume and diversity of citizen-generated feedback. This trend emphasizes the need for classification systems that are not only tailored to specific administrative domains but also robust to the linguistic, contextual, and structural variability inherent in user-submitted content. This study investigates the comparative effectiveness of large language models (LLMs) alongside instruction-tuned embedding models in the task of categorizing public transportation complaints. LLMs were tested using a few-shot inference, where classification is guided by a small set of in-context examples. Embedding models were assessed under three paradigms: label-only zero-shot classification, instruction-based classification, and supervised fine-tuning. Results indicate that fine-tuned embeddings can achieve or exceed the accuracy of LLMs, reaching up to 90 percent, while offering significant reductions in inference latency and computational overhead. E5 embeddings showed consistent generalization across unseen categories and input shifts, whereas BGE-M3 demonstrated measurable gains when adapted to task-specific distributions. Instruction-based classification produced lower accuracy for both models, highlighting the limitations of prompt conditioning in isolation. These findings position multilingual embedding models as a viable alternative to LLMs for classification at scale in data-intensive public sector environments.