Abstract. The research is devoted to solving the urgent problem of modernization of technological cycles at mining and processing plants. The main focus of the article is on automation of the primary stage of separation of fossil raw materials. Today, in many enterprises, quality control of rock mass is based on visual inspection and is carried out by the operator. It is generally believed that with this approach, the human factor introduces subjectivity into the assessment and reduces the accuracy of impurity fixation. That is why the article explores the possibility of implementing computer vision systems for operational sorting.
The research focuses on the development and testing of a binary classification method for digital images, which makes it possible to effectively separate streams into the target product (coal) and waste rock. In the framework of this work, the Random Forest algorithm was chosen as an architectural solution, the hyperparameters of which were optimized by the lattice search method. During the preliminary tests, the algorithm showed stable results in dusty conditions and changing lighting. To train and test the model, a data set of 4027 images of the mountain range was collected. The experiment was based on a comparative analysis of the proposed method with the methods of convolutional neural network (CNN), logistic regression and decision tree. The results confirmed the potential of this method. The model achieved a classification accuracy of 96.5% with an F1-score of 0.896 and a coal detection completeness of 85.7%. It has been found that with accuracy comparable to convolutional networks, the chosen algorithm has an advantage in resource efficiency and the ability to work on Edge devices without a GPU, providing performance of 30-35 FPS. The research results allow us to conclude that the achieved indicators, as well as the stability of the algorithm, make it possible to successfully integrate it into the monitoring system. The proposed solution can become the basis of an autonomous control system at a mining and processing plant without human intervention.
Keywords: computer vision, machine learning, Random Forest, rock classification, conveyor automation, coal industry.
Abstract. This paper examines an approach to building an event-driven serverless architecture for a distributed information system operating under uneven and peak load conditions. Modern digital services are characterized by sharp fluctuations in the intensity of incoming requests, which requires maintaining operational stability, acceptable response times, and the ability to quickly scale computing resources. Traditional monolithic and container-based solutions in such environments often require upfront capacity reservations or respond to load surges with delays. The goal of this study is to develop and experimentally evaluate an architectural solution in which request processing is organized as a stream of independent events using a serverless computing model. To this end, a formalized event model and input load generation scheme are proposed, enabling the reproduction of both normal operating modes and short-term peak impacts. The effectiveness was assessed based on a series of controlled computational experiments under various load scenarios. The key metrics used were average processing latency, the p95 metric, system throughput, and the error rate during periods of increased activity. The results obtained during the study demonstrate that as the load increases, the increase in latency is manageable, and the system maintains operability during short-term overloads. This allows us to consider the proposed approach as a promising solution for scalable distributed services.
Keywords: event-driven architecture, serverless computing, distributed information systems, scalability, peak load, tail latency, performance.
Abstract. This paper presents an experimental evaluation of the efficiency of the proposed digital signature based on a Verkle tree using the Chinese Remainder Theorem. A software implementation of the algorithms for key generation, signature formation, and verification has been developed. In the proposed scheme, the Verkle tree is used for compact representation of commitments, while the Chinese Remainder Theorem is applied to optimize modular computations and improve the computational efficiency of signing and verification operations. An analysis of the time characteristics of the algorithms was carried out, and complexity indicators were obtained. Experimental results were obtained on a fixed computing platform with multiple test runs to ensure statistical reliability. A comparative analysis was performed with a digital signature based on the classical polynomial commitment scheme Kate–Zaverucha–Goldberg (KZG) in terms of the main signature parameters and execution time. The obtained results demonstrate the potential of using the Verkle tree in combination with the Chinese Remainder Theorem for constructing compact and computationally efficient digital signatures.
Keywords: Verkle tree, vector commitment, polynomial commitment, Chinese Remainder Theorem, digital signature, authentication, verification.
Abstract. In recent years, the problem of air pollution has become more and more acute, especially for industrial regions. The constant growth of environmental monitoring data requires not only their accumulation, but also effective intelligent processing. One of the key tasks is the timely detection of abnormal values that can indicate both real emissions of pollutants and errors in measuring systems.
In this paper, an algorithm for detecting anomalies in the atmospheric air monitoring system is proposed, based on a combination of statistical methods and machine learning algorithms. This approach allows you to take into account both simple emissions and more complex, hidden patterns in the data. For primary filtration, the methods of Z-score and interquartile range (IQR) were used, and for a more in-depth analysis, the Isolation Forest algorithm was used, which is able to effectively work with multidimensional ecological time series. The novelty of the study lies in the hybrid decision procedure that combines statistical filtering, unsupervised anomaly detection and meteorological-context interpretation for industrial air pollution monitoring data.
Particular attention is paid to the construction of the system architecture, which is implemented using cloud technologies. This provides the ability to process large amounts of data coming from monitoring sensors, as well as analyze them in near real time. The algorithm was tested on data from the city of Ust-Kamenogorsk, including indicators of the concentration of the main pollutants and meteorological parameters. The results showed that the proposed hybrid approach achieved higher performance than individual methods, reaching Precision = 0.94, Recall = 0.91 and F1-score = 0.92. At the same time, the system is able to automatically record sharp deviations associated with industrial emissions, weather conditions or technical failures.
The practical significance of the work lies in the possibility of introducing the proposed algorithm into environmental information systems and smart city solutions. Its application makes it possible to improve the quality of monitoring, the efficiency of response and the validity of management decisions in the field of environmental protection.
Keywords: atmospheric air monitoring, data anomalies, machine learning, isolation forest, environmental monitoring, cloud technologies.
Abstract. Relation Extraction is a fundamental bridge between unstructured text and formal knowledge representations. Its development for the Kazakh language has been hindered by the scarcity of high-quality annotated semantic resources. While the KazNERD dataset established a robust baseline for entity identification, the transition toward modeling complex interactions between entities remains a critical challenge. This study addresses this bottleneck by introducing the Kazakh Relation Extraction Dataset (KRED), a high-fidelity benchmark constructed through a scalable pipeline that leverages the synergistic capabilities of Large Language Models (LLMs) and human expertise. The annotation workflow used the KazNERD corpus’s verified entity boundaries as a structural base. It included candidate pair generation, zero-shot prompting using GPT-4o-mini, and iterative semantic refinement. Schema-driven normalization and targeted re-annotation with Gemini-3-flash, followed by manual verification, ensured linguistic accuracy. The resulting KRED dataset contains 16,149 relation instances across ten semantic categories. Experiments using transformer architectures such as multilingual BERT, XLM-RoBERTa, and Kaz-RoBERTa show the dataset’s effectiveness. Multilingual BERT performed best, achieving a micro-F1 score of 0.8832 and a macro-F1 score of 0.8113, which provides a solid baseline for future work. This hybrid approach, which uses LLMs, offers a cost-effective alternative to manual labeling. It provides a methodological framework for quickly expanding information extraction resources in low-resource and Turkic languages.
Keywords: natural language processing, information retrieval, low-resource languages, large language models, dataset construction, relation extraction.
Abstract. The article is devoted to the study of the application of machine learning algorithms for the classification of regions of Kazakhstan using demographic data for 2024. The study considers the Decision Tree, Random Forest and k-Nearest Neighbors (KNN) algorithms. They demonstrate high efficiency in solving this problem. Data preprocessing included the calculation of the urban population ratio (urban_ratio), which was used to construct the binary target variable. All three evaluated algorithms demonstrated high performance under the reported experimental settings. The observed differences between the models were limited, with Decision Tree, Random Forest, and KNN showing comparably strong results across the tested partitions. The results indicate the potential of machine learning methods for territorial classification based on demographic indicators; however, the findings should be interpreted with regard to the selected feature set and target construction. In addition, K-means clustering and principal component analysis identified three distinct demographic profiles among the districts, providing a clearer understanding of regional differences.
Keywords: machine learning, classification, demographics, urbanization, Principal Component Analysis, clustering.
Abstract. Tuberculosis remains one of the most widespread infectious diseases, requiring timely and accurate diagnosis. In this regard, this study presents the development and comparative evaluation of deep learning models for the automatic detection of tuberculosis in chest X-ray images. The aim of the study is to analyze the effectiveness of YOLOv8 and RT-DETR model families for detecting pathological changes and to determine the most accurate and robust architecture. The initial dataset consisting of 2441 images was expanded to 5859 using data augmentation techniques and annotated with bounding boxes in the Roboflow environment.
Within the study, YOLOv8 and RT-DETR models were trained and comparatively analyzed. Model performance was evaluated using Precision, Recall, F1-score, and mAP metrics. Experimental results showed that the YOLOv8m model achieved the best performance (Precision = 0.94, Recall = 0.91, mAP50= 0.95, mAP50-95 = 0.72), providing an optimal balance between accuracy and computational efficiency. Based on this model, an intelligent agent for automatic analysis of medical images was developed. The obtained results confirm the effectiveness of YOLO-based models for computer-aided diagnosis and clinical decision support systems.
Keywords: tuberculosis, chest X-ray, deep learning, YOLO, RT-DETR, object detection, medical images, AI agent.
Аbstract. Ensuring flight safety and improving the economic efficiency of aircraft maintenance require intelligent analysis of unstructured textual reports. Traditional topic modelling methods are limited by the loss of semantic context in short messages and the high labour intensity of manual text preprocessing. The subject of this study is methods for the automated extraction of knowledge from technical reports. To achieve this objective, the tasks of designing an optimal computational pipeline, implementing it in software, and conducting experimental validation using real-world data were addressed. The aim of the study is to develop, theoretically substantiate, and experimentally verify a comprehensive method for the automatic identification and detailed interpretation of latent fault subgroups. The proposed method is based on the integration of a transformer model (paraphrase-multilingual-MiniLM-L12-v2) for generating contextual embeddings, the UMAP nonlinear dimensionality reduction algorithm, and HDBSCAN hierarchical density-based clustering. Topic interpretation and the automatic generation of human-readable topic labels are implemented using the c-TF-IDF algorithm, KeyBERTInspired semantic selection, and a locally deployed large language model, Qwen3.5-4B-Instruct. Experimental validation of the method was performed on a dataset comprising 1,971 textual records related to passenger equipment collected over a seven-year operational period from nine aircraft. The proposed method successfully identified seven stable and meaningful defect categories while isolating 11.7% of atypical records into a noise cluster. Comparison with a random baseline confirmed the high statistical significance of the results. The topic diversity score reached 0.8286, which falls within the recommended range. The proposed method outperforms the classical LDA model in terms of the component-wise purity of the extracted topics and eliminates the need for complex manual data preprocessing. The solution is recommended for scaling to other sections of the ATA 100 standard and for integration into intelligent decision-support systems of aviation enterprises to optimise maintenance planning and component inventory management.
Keywords: maintenance, air transport, text mining, contextual embeddings, density-based clustering, topic modeling, large language models.
Abstract. The incidence of cancer currently remains one of the most significant problems of modern society. Lung cancer is one of the most lethal forms of oncopathology, which is largely due to its detection in the late stages of the tumor process. The effectiveness of screening and early diagnosis programs has a direct impact on disease prognosis and patient survival rates. In this regard, special attention has been paid in recent years to the development of intelligent expert systems based on deep learning methods, including convolutional neural networks (CNN), transformer architectures and their hybrid solutions. This paper considers the problem of early diagnosis of lung cancer using deep learning neural network models. During the research, models were developed and analyzed using the following machine learning methods: logistic regression, random forest classifier, support vector machine (SVM), extreme tree classifier, XGBoost, CatBoost, gradient boosting and multilayer perceptron (MLP). The indicators of hereditary predisposition, the patient’s gender and smoking experience were used as input factors to the neural network model. The best results were demonstrated by a model based on a multilayer perceptron (MLPClassifier), which reached the maximum value of the ROC-AUC metric, equal to 0.9405, with an accuracy level of over 93%. The obtained indicators indicate the high ability of the model to correctly rank patients by risk groups. The randomForest and SVM algorithms, which took the second position in classification quality, showed comparable results. The developed neural network model makes it possible to assess the likelihood of developing lung cancer, as well as to make recommendations regarding the possible presence of concomitant diseases such as disseminated pulmonary tuberculosis, sarcoidosis, pneumonia and pulmonary fibrosis. In the future, it is planned to expand the functionality of the system by integrating a medical image recognition module, which will create a comprehensive solution for early diagnosis of lung cancer.
Keywords: information systems, artificial intelligence, deep learning methods, intelligent data analysis, intelligent system, databases, convolutional neural network, transformers, hybrid models, explicable AI.
Abstract. The rapid development of the Internet of Things (IoT) is accompanied by a growing number of connected devices and an increasing volume of cyberattacks, which necessitates the development of effective and scalable protection mechanisms. This paper proposes IoTector, a platform for network traffic monitoring and cyberattack detection in IoT networks. IoTector is designed as an intelligent gateway deployed between IoT devices and the network infrastructure, providing real-time detection of attacks and anomalies. To analyze network traffic, the system employs deep learning models, including DNN, CNN, and CNN–BiLSTM, enabling effective detection of various attack types. A prototype of the platform was implemented on Raspberry Pi 5 and supports device connectivity through wireless technologies such as Wi-Fi and Bluetooth. In addition, a software interface was developed to provide network status monitoring, threat visualization, device management, and support for model training. Unlike traditional intrusion detection systems mainly focused on centralized traffic analysis, the proposed approach integrates intelligent filtering, monitoring, and management within a unified platform. The obtained results confirm the high effectiveness of IoTector in detecting attacks and anomalies, as well as its practical applicability in real-world IoT environments.
Keywords: Internet of Things, IoT security, intrusion detection system, IoTector, federated learning, deep learning, network traffic monitoring, distributed systems.
Abstract. The paper addresses the problem of rapid response to flood events, which frequently occur in the territory of the Republic of Kazakhstan. The relevance of the study is determined by the lack of reliable and timely spatial data required for early flood detection and monitoring. The aim of the research is to develop and evaluate a crowdsourcing-based mobile information system for flood monitoring using data from the Global Navigation Satellite System (GNSS). The object of the study is the process of collecting and processing spatial flood data, while the subject of the research focuses on mobile data collection methods based on the citizen sensing concept. The methodology includes the design and development of a mobile application for the Android operating system, integration with the Global Navigation Satellite System Application Programming Interface (GNSS API), and photo documentation using standardized image metadata (Exchangeable Image File Format, EXIF). As a result, an integrated architecture of a flood monitoring information system has been developed. The proposed architecture consists of a mobile application, a server-side module for data storage and processing, and a web-based control panel for emergency management services. The mobile application provides automatic coordinate acquisition with an accuracy of 8–12 m, geotagging of photographic data, and input of contextual information such as water level and terrain type. The results of a local pilot experiment demonstrate that the proposed system is practically applicable for early flood detection and supports timely decision-making and emergency response under the conditions of Kazakhstan.
Keywords: Global Navigation Satellite System, crowdsourcing method, mobile system, citizen sensing, flood monitoring, information system architecture.
Abstract. The Kazakh language is designed so that grammatical information is distributed along a chain of affixes, rather than concentrated in individual words. This in itself creates difficulties for automated systems. The situation is further complicated by the constant influx of Russian and English borrowings, which are partially dislodged from the phonological logic of the language, and as a result, the morphological analyzer is faced with not one type of difficult forms, but with several fundamentally different ones. Existing systems evaluate the performance of NLP models in an aggregated manner, without distinguishing the types of anomalies by the mechanism of their impact on the analysis, which does not allow targeted management of processing. The subject of this study is the relationship between the types of morphological errors and performance indicators of transformative NLP models for the Kazakh language. The aim of the work is to develop a formalized taxonomy of morphological anomalies and integrate it into a hybrid NLP architecture as a routing mechanism for analytical components. The study uses formal linguistic analysis, the corpus method based on the Universal Dependencies Kazakh-KTB (1,047 sentences), computational modeling and ablative analysis. The KazMorphCorpus-2026 hybrid architecture combines rule-based FST analysis, CRF disambiguation, the KazRoBERTa transformer module and the MFRN feature morphological compatibility verification module. Based on the results of the study, a five–class taxonomy of morphological anomalies was proposed – borrowings (BOR), affixal complexity (AFC), segmentation disorders (SEG), conflicts of grammatical features (AGR) and neutral class (NONE), integrated into the system as a control routing mechanism. In the test sample, the system reaches Accuracy = 87.4% and Macro-F1 = 0.86; the largest increase in quality was recorded for the AGR (ΔF1 = +0.14) and AFC (ΔF1 = +0.12) classes. The experiment confirmed that different types of morphological anomalies have different effects on the operation of the transformer model, and this difference is of practical importance. Systems that diagnose the type of anomaly prior to analysis and direct the token to a suitable component produce a result that is easier to interpret and easier to improve purposefully.
Keywords: morphological error taxonomy, morphological routing, hybrid NLP architecture, transformer models, KazRoBERTa, FST, CRF, MFRN, agglutinative language.