Search

Search Results

• TWO-STAGE BOOSTING OF BINARY CLASSIFICATION BASED ON THE APPLICATION OF BIOINSPIRED ALGORITHMS

  D. V. Balabanov , A. V. Kovtun , Y. A. Kravchenko

  2020-10-11

  Abstract ▼

  In the process of solving a wide range of applied problems, it becomes necessary to decompose
  objects. As a result, the classification problem is an urgent problem in modern data mining
  systems. Binary classification is one of the most important tasks, and has a number of unsolved
  problems. One such problem is the effectiveness of automated classification. In the tasks of automated
  classification, it is relevant to use the algorithmic apparatus of evolutionary computing.
  Thus, it is advisable to use genetic and bio-inspired algorithms in the task of finding the optimalvalues of the classifier parameters. To solve this problem, it is proposed to apply the particle
  swarm algorithm (PSO). This algorithm in the context of the task of finding suboptimal values of
  the parameters of the classifier is able to provide high quality classification. A modification of the
  algorithm is a dynamic change in the coordinate values that are responsible for the type of kernel
  function. This revision can significantly reduce the time spent developing the classifier. To increase
  the classification efficiency, it is advisable to use ensembles of algorithms. The paper presents
  the structure of a two-level classifier. At the first level of this classifier, an ensemble of simple
  classifiers is formed that form the training set, which is further used by the particle swarm
  algorithm in the second stage. This approach can significantly reduce time costs, as well as improve
  the quality of the resulting solutions. The particle swarm algorithm (PSO), in the context of
  the task of finding suboptimal values of the parameters of the classifier, is able to provide high
  quality classification. The proposed two-level algorithm has been experimentally tested. A comparison
  is made with analogues, comparative charts are given. The described studies show that
  the work is of high theoretical significance, and the conducted experimental studies prove high
  practical significance.

• CLASSIFIER OF IMAGES OF AGRICULTURAL CROPS SEEDS USING A CONVOLUTION NEURAL NETWORK

  V. A. Derkachev, V. V. Bakhchevnikov, A. N. Bakumenko

  2020-11-22

  Abstract ▼

  This article discusses the creation of a convolutional neural network architecture that classifies
  images of crops (in particular wheat) for subsequent use in an optical seed separator (photo
  separator). Interest in the design of neural networks for classifying images has recently increased
  significantly, which is associated both with the development of the theory of deep neural networks
  and the increased computing power of desktop computers, as well as the transfer of computing to
  graphic processors. The aim of the article is to develop the architecture of a neural network that
  allows the separation of the input flow of wheat seeds into two classes: “good” seeds and “bad”
  (with defects in shape and color) seeds. The architecture of the resulting neural network is convolutional,
  because, unlike a fully connected one, this class of neural networks is within certain limits
  immune to changes in the scale and angle of rotation of objects in the input data. In the work,
  for the formation of training, validation and test samples, seed images obtained using a household
  camera were used, which negatively affected the results of training and testing the neural network
  regarding the possible result of application in a real photo separator. The architecture of the developed
  neural network is preliminarily optimized for use on FPGAs, however, in the considered
  case, the transition from the values of weighting factors from the data type from a floating point to
  an integer type has not been made, which can lead to a decrease in the accuracy of the neural
  network, while significantly reducing the amount of resources FPGA. Application of the proposed
  architecture allows one to obtain a fairly accurate estimate of classified wheat seeds from verification
  and test data sets.

• DEVELOPMENT OF A METHOD FOR PERSONAL IDENTIFICATION BASED ON THE PATTERN OF PALM VEINS

  V.А. Chastikova, S.А. Zherlitsyn

  2022-11-01

  Abstract ▼

  The article describes the work on the creation of a neural network method for identifying
  a person based on the mechanism of scanning and analyzing the pattern of palm veins as a biometric
  parameter. As part of the study, the prerequisites, goals and reasons for which the deve lopment
  of a reliable biometric identification system is an important and relevant area of activity
  are described. A number of problems are formulated that are inherent in existing methods for
  solving the problem: the graph method and the method based on calculating the distance expressed
  in various interval metrics. The description of the principles of their work is given.
  The tasks solved by personal identification systems are formulated: comparison of the subject of
  identification with its identifier, which uniquely identifies this subject in the information system.
  A mechanism for reading a pattern of veins from the palm of the hand, developed for analyzing
  an image obtained with a digital camera sensitive to infrared radiation, is described. When the
  palm is in the frame, illuminated by the light of the near infrared range, the image obtained
  from the camera becomes noticeable pattern of veins, vessels and capillaries that lie under the
  skin. Depending on the organization, the identification system may, based on the provided identifier,
  determine the appropriate access subject or verify that the same identifier belongs to the
  intended subject. Three methods for further analysis of biometric data and personal identification
  are given: approaches based on categorical classification and binary classification, as well
  as a combined approach, in which identification is first used by the first method, and then, by
  the second, but already for a known access identifier defined on the first stage. The resulting
  architecture of the neural network for the categorical classification of the vein pattern is pr esented,
  a method for calculating the number of model parameters depending on the number of
  registered subjects is described. The main conclusions and experimental measurements of the
  accuracy of the system when implementing various methods are presented, as well as diagrams of
  changes in the accuracy of models during training. The main advantages and disadvantages of the
  above methods are revealed.

• APPLICATION OF CONVOLUTIONAL NEURAL NETWORKS FOR TECHNICAL OBJECT RECOGNITION IN THE INTERESTS OF RADIO MONITORING

  D. V. Shumkov, I.V. Titkov, P.А. Gulevich

  2025-04-27

  Abstract ▼

  The article examines the possibility of using convolutional neural networks for technical object
  recognition in the context of radio monitoring. The focus is on the development and optimization of algorithms
  for processing radar signals using deep neural networks. Studies have shown that the use of CNN
  can significantly improve the classification accuracy of radio signals compared to traditional processingmethods. The developed approach is based on the extraction of hierarchical features from spectral images
  of radio signals and their subsequent classification using a trained neural network. The paper presents the
  results of experimental studies conducted on a dataset of more than 10,000 samples of radio signals of
  various types. It is shown that the proposed technique ensures recognition accuracy of up to 94% when
  working with noisy signals and the probability of a false alarm is no more than 0.05. Special attention is
  paid to the choice of neural network architecture for the specifics of the radio monitoring task. The options
  for converting radio signals into a spectral image for real-time processing were also considered in
  detail. Data preprocessing methods have been developed, including amplitude normalization, frequency
  correction, and interference elimination. The results of the study can be used in radio broadcast control
  systems and to ensure electromagnetic compatibility of electronic devices. The results obtained demonstrate
  the prospects of using CNN in the tasks of technical recognition of radio monitoring objects and
  open new opportunities for the development of intelligent radar information processing methods. Promising
  areas of further research include the development of adaptive neural network training methods in a
  changing radio environment and the creation of hybrid systems combining traditional signal processing
  methods with modern neural network algorithms

• MULTIMODAL DATA FEATURE EXTRACTION METHOD FOR NETWORK ATTACK CLASSIFICATION

  A.V. Balyberdin

  6-16

  2025-07-24

  Abstract ▼

  An intrusion detection system (IDS) is an important component of corporate data network (CDN) protection. IDS analyzes network traffic and detects network attacks. Depending on the detection methods, IDS can be classified into the following types of systems: signature-based analysis systems, anomaly detection systems (ADS), and hybrid systems combining the aforementioned approaches. Recently, anomaly detection systems (IDS) have been actively developing. For anomaly detection systems, network attacks are anomalous behavior of network traffic consisting of a set of features or event attributes. Modern IDS are based on machine and deep learning methods, and therefore the detection of network attacks and anomalies is formulated as a classification and clustering problem. To solve these problems, methods for optimizing the feature space of network traffic are required. The aim of the work is to develop a feature extraction method based on a multimodal approach to representing network traffic data for classifying network attacks. The paper considers the analysis of relevant studies on feature extraction methods from various fields. The objective of the study is to improve classification efficiency using a multimodal representation of network traffic features. The result of the work is a method for extracting data features based on two modalities: a spectral representation of network traffic features and an image feature matrix. The novelty of the presented method lies in the application of the windowed Fourier transform method for network traffic events, followed by the calculation of spectral features for discrete signals, as well as the transformation of data features into an image matrix and its expansion to optimize the feature space using a convolutional neural network (CNN). Evaluation of the multimodal method showed that this method increased the classification accuracy for unbalanced classes of network attacks

• DEVELOPMENT OF A CHATBOT FOR CLASSIFICATION AND ANALYSIS OF NATURAL LANGUAGE TEXTS USING LOCAL LARGE LANGUAGE MODELS

  Juman Hussain Mohammad , Juman Hussain Mohammad , Y.А. Kravchenko

  159-171

  2025-07-24

  Abstract ▼

  This paper explores local large language models (LLMs) and their application in text classification tasks, while also comparing their performance with traditional methods. The paper provides a comprehensive review of several key local LLMs, with particular focus on their architectural advantages, characteristics, and application domains. Specifically, we examine models with varying numbers of parameters, their ability to adapt to specialized domains, and their computational requirements when deployed on local hardware. Special emphasis is placed on the trade-offs between performance and resource efficiency. As a practical contribution, we developed a chatbot that utilizes local LLMs (such as DeepSeek, Gemma, and Llama2 via Ollama) to classify incoming texts into predefined categories, demonstrating the operation of these models without cloud computing. The system features a modular architecture that allows for easy integration of new models and comparison of their effectiveness. The computational experiment involves evaluating the accuracy and inference speed of local LLMs compared to simpler methods such as Sentence-BERT, TF-IDF and BoWC, highlighting scenarios in which local models outperform or underperform traditional approaches. Testing was conducted using the benchmark BBC dataset. The results show that language models (including 7-billion parameter models) demonstrate strong and logically consistent classification performance in natural language text processing. However, their results are not perfect for benchmark datasets. Notably, we identified cases where all tested models, including traditional methods, misclassified documents, suggesting potential issues with data labeling. These findings indicate the need to reconsider benchmark labels in standard datasets, particularly for domains with subjective categories where expert evaluations may vary significantly. On the other hand, while local LLMs lag behind cloud-based solutions in speed, their advantages in data privacy and offline operation make them suitable for specialized tasks. This is particularly valuable in medical and financial institutions where protection of sensitive information is critical, and where local models can be fine-tuned for specific business processes without the constraints of cloud APIs.

• FAILURE PREDICTION USING FACTOR ANALYSIS METHODS

  Е.S. Podoplelova

  213-223

  2025-07-24

  Abstract ▼

  This article discusses the application of a risk assessment method based on the combination of the FMEA (failure mode and effect analysis) methodology and the MCDM (Multiple Criteria Decision Making) methods. This approach allows taking into account both expert knowledge and historical data on the operation of the equipment. MCDM methods process the assessment more flexibly in comparison with the standard method of calculating the priority number of risks (PRN), which helps to better assess the risks by three criteria: the probability of occurrence, the complexity of detection and the severity of the consequences. One of the criteria can be obtained not only through an expert assessment, but also on the basis of data recording the operation of the equipment. This approach was tested using the example of synthetic open-source data on the operating modes of production equipment. The task was to predict both the failure itself and its type, as well as to identify the factors that have the greatest impact on the failure. For this purpose, data preprocessing was carried out, during which it was necessary to eliminate the imbalance of classes. There are several approaches to solving this problem, aimed at reducing the dominant class or generating instances of poorly represented classes. In this example, random reduction of the number of records without errors was used. Then, AdaBoost, Random Forest and LinearSVC were compared as classification algorithms. Since multi-class classification was required, it was decided to use the one-vs-the-rest strategy. As a result, it was possible to achieve 86% forecasting accuracy by F-measure using the AdaBoost and Random Forest algorithms. LinearSVC turned out to be ineffective. Thus, the resulting forecasting model recognizes different types of errors, but there is room for improvement, which requires a larger sample, including more examples with different types of failure. Based on this, this approach as an alternative to expert assessment is promising, improving objectivity, and also making it possible to foresee risks and prevent a real failure or risk-related incident.

• ANALYTICAL REVIEW OF THE DECISION TREE ALGORITHM IN DATA INTELLIGENCE TECHNOLOGY

  E.V. Kuliev, V.A. Semenov, A.V. Kotelva, S.V. Ignateva

  2022-05-26

  Abstract ▼

  The decision algorithm is the preferred filtering algorithm in data mining technology, and
  its results are usually chosen in the form of "if-then" rules. Algorithm C4.5 is one of the decision
  algorithms that takes advantage of the ease of understanding and increasing importance, and also
  takes advantage of the advanced information rate gain of its advanced ID3 algorithm. After the
  theoretical analysis of the information, the algorithm C4.5 is selected to analyze the results of
  performance appraisal, and enterprise performance appraisal decisions by collecting data, preprocessing
  data, calculating information gain and determining selection parameters. The system isdeveloped in B/S architecture, an R&D project management platform that can perform evaluation
  analysis with decision analysis results evaluation tools and web coverage. The system includes
  information storage, task management, reporting, receipt and presentation control, information
  visualization and other functions of the management information system functions. They can realize
  project management functions, such as creating and managing a project, flow tasks, filling and
  managing information about functions, creating a performance evaluation system, creating reports
  of various sizes, building management. decision decision algorithm as the core technology,
  the system acquires scientific significant project management information with high data accuracy,
  and realizes visualization, which can help the enterprise to have a good management system in
  large areas. Task management, reporting, audit control, information visualization and other functions
  of the system's management reporting management functions are included.

• TEXT VECTORIZATION USING DATA MINING METHODS

  Ali Mahmoud Mansour , Juman Hussain Mohammad, Y. A. Kravchenko

  2021-07-18

  Abstract ▼

  In the text mining tasks, textual representation should be not only efficient but also interpretable,
  as this enables an understanding of the operational logic underlying the data mining
  models. Traditional text vectorization methods such as TF-IDF and bag-of-words are effective and
  characterized by intuitive interpretability, but suffer from the «curse of dimensionality», and they
  are unable to capture the meanings of words. On the other hand, modern distributed methods effectively
  capture the hidden semantics, but they are computationally intensive, time-consuming,
  and uninterpretable. This article proposes a new text vectorization method called Bag of weighted
  Concepts BoWC that presents a document according to the concepts’ information it contains. The
  proposed method creates concepts by clustering word vectors (i.e. word embedding) then uses the
  frequencies of these concept clusters to represent document vectors. To enrich the resulted document
  representation, a new modified weighting function is proposed for weighting concepts based
  on statistics extracted from word embedding information. The generated vectors are characterized
  by interpretability, low dimensionality, high accuracy, and low computational costs when used in
  data mining tasks. The proposed method has been tested on five different benchmark datasets in
  two data mining tasks; document clustering and classification, and compared with several baselines,
  including Bag-of-words, TF-IDF, Averaged GloVe, Bag-of-Concepts, and VLAC. The results
  indicate that BoWC outperforms most baselines and gives 7 % better accuracy on average

• POPULATION ALGORITHM FOR CONSTRUCTING A TREE OF SOLUTIONS BY METHOD OF CRYSTALLIZATION OF ALTERNATIVES FIELD

  B.K. Lebedev , O.B. Lebedev , V. B. Lebedev

  2020-11-22

  Abstract ▼

  In some cases, it becomes necessary to establish a correspondence between the declared
  and actual value of a categorical variable on the basis of a set of object characteristics. In this
  case, there is a need for a classifier with an optimal sequence of the considered attributes with agiven value of the objective function. The target variable can be: yes, no, variety number, class
  number, etc. This paper solves the problem of constructing a classification model in the form of an
  optimal sequence of the considered attributes and their values included in the route from the root
  vertex to the terminal vertex with a given value of the target variable. If a classifier is required
  that includes the possibility of alternative answers, then first, independently from each other, optimal
  routes are built for each value of the target variable, and then these routes are combined
  ("glued") into a single binary decision tree. In the algorithm for constructing a classifier based on
  the method of crystallization of a placer of alternatives, each solution Qk is interpreted as an oriented
  route Mk on a binary decision tree. Let us call the ordinal number of an element in the directed
  route Mk the position siS={si|i=1,2,…,nA}. An element of the route Mk is the pair (xi, ui-),
  where xi corresponds to Ai. ui- in the route Mk is an edge outgoing from xi and corresponds to the
  value Ai chosen together with Ai. The second index of the element ui- is determined after the choice
  of Ai, placed in the position sj+1 adjacent to sj. The work of the decision tree construction algorithm
  is based on the use of collective evolutionary memory, which is understood as information
  reflecting the history of the search for a solution. The algorithm takes into account the tendency to
  use alternatives from the best solutions found. The peculiarities are the presence of an indirect
  exchange of information – stigmerges. The totality of data on alternatives and their assessments
  constitutes a scattering of alternatives. The key points of the analysis of alternatives in the process
  of evolutionary collective adaptation are considered. Experimental studies have shown that the
  developed algorithm finds solutions that are not inferior in quality, and sometimes surpass their
  counterparts by an average of 3–4 %. The time complexity of the algorithm, obtained experimentally,
  lies within O(n2)-O(n3).

• COMBINING SEGMENTATION, TRACKING, AND CLASSIFICATION MODELS TO SOLVE VIDEO ANALYTICS PROBLEMS

  V.D. Matveev, А. Е. Arkhipov, I. S. Fomin

  2025-04-27

  Abstract ▼

  The task of detecting obstacles in front of a mobile robot has been successfully solved long ago using
  laser and ultrasonic sensors. However, obstacles that are not detected by these types of sensors may endanger
  the safety of the robot. To detect them in the work, it is proposed to use a technical vision system
  (STZ), the information from which is processed by a semantic segmentation neural network, which returns
  the mask of the obstacle on the frame and its class. The basis for such a network was the SAM universal segmentation
  network, which requires further development to be applied to the semantic segmentation task.
  The peculiarity of this network is its universal applicability, that is, the ability to select any objects in any
  filming situation. At the same time, SAM does not predict the semantics of the object. In this paper, an additional
  module is proposed that makes it possible to implement semantic segmentation by classifying the features
  of the selected objects. The possibility of using such a module to solve the problem of supplementing the
  network output with new information is substantiated. The classification result is then fed into the same filtering
  algorithm as the masks to ensure consistency between the result of the universal network and the complementary
  module. After integrating the module with the model, a new semantic segmentation model was
  obtained, called RTC-SAM in the work. It was used to perform semantic segmentation of a publicly available
  dataset with images of an open area. The 45% result obtained by the IoU metric exceeds the result of existing
  methods by 13%. The images of the results of using the new network shown in the work make it possible to
  verify its performance. It also describes the testing of the developed solution with a study of the performance
  of the developed model on a PC and a mobile computer. The algorithm on the mobile computer shows insufficient
  speed to enter real-time mode – more than 3.5 seconds to process one frame. In this regard, one of
  the directions of further research in the field of improving system performance.

• MONITORING OF THE EDUCATION QUALITY AND IMPLEMENTING OF INDIVIDUAL LEARNING: DEMONSTRATION OF APPROACHES AND EDUCATIONAL DATA MINING ALGORITHMS

  Yass Khudheir Salal , S. M. Abdullaev

  2020-10-11

  Abstract ▼

  The quality monitoring system for traditional and distance education requires the development
  of machine learning classification and quantification techniques necessary to predict individual
  and collective student performance. This article theoretically and experimentally shows that
  the most promising approach that simultaneously solves both forecast tasks is to create heterogeneous
  ensembles consisting of an odd number of different base classifiers, such as decision trees,
  simple neural networks, naive Bayesian classifier and others. By training and testing 11 different
  binary classifiers on six different samples of educational data, we show that the individual determined
  forecast of such ensembles exceeds the accuracy of forecasts of both individual base classifiers
  and homogeneous ensembles created by bagging and busting technologies. The advantage of
  heterogeneous ensembles is decisive when we deal with the imbalance of sample characteristic ofeducational data. In these cases, only the forecasts with accuracies exceeding the relative frequency
  of the class of objects dominating in the sample of data can be considered as useful forecasts.
  The main advantage of the heterogeneous ensemble is the ability to transform the deterministic
  forecast into a probabilistic forecast, when instead of referring the object to a particular class, the
  probability of its belonging to individual classes is given. On this basis, we have proposed a new
  method of binary quantification, where individual probabilities of belonging to each of the classes
  of objects are summed up separately, and the resulting total probabilities are interpreted as relative
  frequencies of objects in the sample. As a result of experiments, it is shown that such ensemble
  binary quantification is significantly superior to the traditional "classify and count" method.

• LULC-ANALYSIS OF LAND-USE WITH THE HELP OF UNSUPERVISED CLASSIFICATION

  Ranjana Waman Gore , Ratnadeep R. Deshmukh, Priyanka U. Randive, Mishra Abhilasha , I. B. Abbasov

  2020-10-11

  Abstract ▼

  Land-use and vegetation cover are the natural state of the earth's surface. Remote sensing is a
  very important land use study (LULC) method. Various classification methods are used to analyze land
  cover in remote sensing. These methods do not require prior information on land cover or land use
  types. Two classification methods are most commonly used to analyze remote sensing images. These
  include controlled classification and uncontrolled classification. The objectives of the proposed work
  are to use unsupervised classification methods to find clusters, determine land use types, and compare
  these methods with interactive analysis of self-organization data (ISODATA). Hyperion sensor images
  were used for land use analysis. The Hyperion sensor has two hundred and forty-two bands, but fewbands provide useful information for spectral analysis. Therefore, bands that do not contain useful information
  are identified and removed. After processing the input image according to this algorithm, out
  of 242 bands, only one hundred and sixty-five bands remain. This takes into account radiometric calibration
  and an important correction of atmospheric factors. Then, based on the results of processing
  using the proposed methods, clusters are formed to study land use using a hyperspectral image. To form
  clusters, the pixels were grouped based on the selected data. Pixels from the same cluster have more
  similarity, while pixels from different clusters differ from each other. Based on the results, it is concluded
  that the clustering method (k-means) allows better identification or prediction of land use based on a
  high-resolution hyperspectral image than the Interactive Self-Organization Data Analysis (ISODATA)
  method. The output image, which is the result of clustering, can be used to identify different types of land
  use objects. The LULC classes predicted are Water Body, Agriculture Land, other Vegetation, Built Up
  or settlement, Bare Land and Rocky region.

• IMAGE MATCHING USING DIFFERENT KEYPOINTS TYPES

  K. I. Morev , A.V. Bozhenyuk

  2020-10-11

  Abstract ▼

  The work is devoted to experiments with various methods of selecting special points on images,
  followed by their description with a binary descriptor and comparison by a full search method.
  This paper actively uses the method of describing the neighborhood of singular points, based
  on the construction of a binary string that characterizes changes in the brightness of pixels in the
  described neighborhood. The resulting string is obtained by comparing the brightness of pixels
  according to a specific template. Today, the use of special points when working with images allows
  you to develop applied methods in various areas of computer vision with increased requirements
  for working time and resistance to sudden changes in scenes. The paper presents the results
  of experiments with special points of various classes, the classification is given in section 1. During
  the experiments, methods implemented in the OpenCV library were used. The paper provides
  brief descriptions of the methods used in experiments. Section 1 of the paper offers a classification
  of modern types of singular points of images and provides a brief description of popular methods
  for detecting the described types of singular points. In section 2, the authors give a General description
  of methods for working with special image points. Section 3 describes the experiments
  that are being carried out with the comparison of special points of different types described by a
  single descriptor, and reveals their results. The experiments performed allow us to identify the
  strengths and weaknesses of bundles of different types of singular points when comparing them.

• FEATURES OF THE FORMATION OF THE PROCESS OF CLASSIFYING THE CONDITION OF A TECHNICAL FACILITY BASED ON THE ANALYSIS OF POINTS IN THE TIME SERIES OF THE PARAMETER

  S.I. Klevtsov

  47-57

  2025-10-01

  Abstract ▼

  Assessment of the operability of a technical facility in real time is important for the stable and trouble-free operation of the facility during its operation. Previously, a classification model for the rate of parameter change was proposed based on specialized point cloud processing of a time series segment without trend extraction. However, some proposals, for example, related to the non-inclusion of some points of the series in the model construction procedure, were not sufficiently justified and are an unobvious attempt to get rid of abnormal values of the time series. Some stages of the model implementation, for example, building an ellipse on a transformed point cloud, require a detailed representation, which is important for further model training and classification.  In the article, as part of the preliminary data preparation, a procedure is proposed for detecting and screening out abnormal values of the time series of a parameter based on a modification of the Irwin method. In addition, an updated scheme for evaluating the values of the criterion in the classification model for the condition of a technical facility parameter is presented. The ellipse compression ratio is used as the evaluation criterion, which is based on a cloud of scatter plot points cut out by a sliding time window from the time series of the parameter. An iterative ellipse construction procedure has been developed for this purpose. The new procedure provides a more informed and accurate assessment of the criterion. Thus, a modified model has been built that will allow real-time assessment of the occurrence of an emergency situation at an early stage of its development.
  The evaluation procedure can be implemented as part of the hardware and software of the monitoring system of a technical facility

• DETECTION OF CYBER INTRUSIONS BASED ON NETWORK TRAFFIC AND USER BEHAVIOR USING THE UNSW-NB15 DATASET

  V. А. Chastikova , К.V. Kozachek , Е.S. Korobskaya , V. P. Kravtsov

  229-243

  2025-11-10

  Abstract ▼

  The article focuses on the study of user behavior and the creation of behavioral models. This helps to improve the accuracy of anomaly detection and quickly identify non-standard network activity.
  The purpose of this study is to compare the effectiveness of two machine learning models – the multilayer perceptron (MLP) and the Random Forest algorithm – for detecting cyber intrusions based on the analysis of network traffic and user behavior. Behavioral models make it possible to detect deviations from normal user activity and network interactions, which significantly increases the completeness of cyber intrusion detection. The study used the UNSW-NB15 dataset, which includes current types of attacks and characteristics of both network traffic and user activity. Prior to the implementation of the models, preliminary data processing, feature selection, normalization and coding of categorical features were carried out.
  The models were evaluated using various metrics such as accuracy, recall, AUC-ROC, precision,
  F1-score, and others. The results of the study showed that the Random Forest algorithm provides high classification accuracy (95%), and the multilayer perceptron (MLP), in turn, achieved outstanding results in AUC (0.9830) and accuracy (precision, 0.9869). The paper presents an analysis and characterization of methods for analyzing user behavior and classifying network traffic, a comparison of data sets for intrusion detection systems (IDS), and practical recommendations for choosing models depending on operating conditions. The results of the study can be useful in the development of adaptive protection systems that combine high accuracy and speed

• RECOGNITION OF EMOTIONAL STATES IN RUSSIAN SPEECH USING MFCC FUNCTIONS AND THE BLSTM MODEL FOR THE DUSHA DATASET

  P.G. Bukina , А.А. Merinov , S.S. Kharchenko , Е.Y. Kostyuchenko

  240-248

  2025-12-30

  Abstract ▼

  This paper investigates the task of automatic emotion recognition from speech signals using contemporary deep learning techniques. The relevance of this study arises from the increasing demand for intelligent systems capable of assessing human emotional states, with potential applications in medicine, psychology, information systems, and personnel management. The primary objective is to develop an efficient neural network model for emotion recognition in Russian speech that outperforms existing state-of-the-art architectures. The experiments were conducted using the open-source Russian-language dataset Dusha, which contains 300,000 audio recordings. A total of 183,055 samples from the Crowd subset, annotated with four emotional categories—joy, sadness, anger, and neutral state—were used for training. Mel-frequency cepstral coefficients (MFCCs) were extracted as input features (20 coefficients with a
  20 ms window and 10 ms overlap), followed by normalization. The baseline architecture employed a bidirectional long short-term memory network (BLSTM), capable of modeling both past and future temporal dependencies. To improve generalization and mitigate overfitting, the model was enhanced with convolutional layers (CNN), MaxPooling layers, and regularization mechanisms including Dropout and Batch Normalization. The resulting hybrid CNN–BLSTM architecture achieved 62.9% accuracy on the test set, exceeding the baseline performance (56.2%) by 6.7%. The results were further compared with state-of-the-art architectures such as MobileNetV2, HuBERT, and WavLM. The analysis highlights future directions for improving model performance through structural optimization, class balancing, and incorporation of additional acoustic features.

• A NEW REPRODUCIBILITY METRIC FOR COMPARING TIME SERIES CLASSIFIERS

  М. О. Dobrokhvalov , А.Y. Filatov , Е.А. Chegodaeva

  2026-02-27

  Abstract ▼

  Experimental reproducibility constitutes a critical cornerstone of modern machine learning research, yet random initialization seed selection substantially influences final model performance, creating challenges for principled comparison of different architectures and methods. Random seed effects on convolutional time series classifiers were quantified, and a principled comparison criterion was established. Two 1D architectures, FCN and ResNet, were trained on seven public datasets containing different data. 55 independent runs for each combination of model and dataset were performed nder controlled pseudorandomness in Python, NumPy, and PyTorch. Deterministic backends were enabled, and identical hyperparameters were used across runs. Normality of seed-wise accuracy distributions was assessed with the Shapiro–Wilk and Anderson–Darling tests. Accuracy variability attributable to seed choice reached up to 12 percentage points in some settings, with magnitude dependent on dataset and architecture. The distributions were found to be non-normal in most cases, indicating that confidence intervals predicated on normality are unreliable. To enable fair comparison across runs, a reproducibility meta-metric, RM, was introduced that subtracts a dispersion penalty from the mean and depends on the number of runs and a tunable coefficient λ. RM was shown to lie between the empirical minimum and the mean, to approach the lower bound for small sample sizes, and to converge toward the mean as the number of runs increases. Portability of the approach was examined on an additional architecture, DenseNet, confirming expected behavior. Practical value is provided by RM metric rankings reflect both performance and stability. In this way, reproducibility and the credibility of empirical conclusions are strengthened

• AN ALGORITHM FOR CONTROLLING AN AUTONOMOUS UNDERWATER VEHICLE WHEN SEARCHING FOR A DESIGNATED BOTTOM OBJECT WITH THE INTEGRATED USE OF VARIOUS BOTTOM MONITORING TOOLS

  V.S. Bykova , А.I. Mashoshin

  2026-04-29

  Abstract ▼

  The search for designated bottom objects is one of the most difficult tasks solved by the AUV, due to a number of factors, the main of which are: the variety of search objects (sunken submarines, surface ships, airplanes, helicopters, mines, underwater pipelines and communication cables, various underwater infrastructure), the need for integrated use for search for various bottom monitoring tools that differ in their physical principles of operation, resolution, and search performance. The purpose of the work, the results of which are presented in the article, was to develop an algorithm for managing the AUV when searching for a designated bottom object that meets these requirements, and to verify it using a digital polygon and a digital twin of the AUV. The probability of correctly attributing the detected bottom object to the search object was chosen as a criterion for choosing a bottom monitoring tool in each specific case. As a result, the logic of searching for a designated bottom object is as follows. The search for bottom objects is carried out using a tool with maximum search performance. When a bottom object is detected, the probability of its belonging to the search object is determined. If it exceeds the set high threshold, a decision is made to locate the designated bottom object. If it is less than the specified low threshold, it is decided that an extraneous object has been detected. In other cases, a decision is made on the need to examine the object with a higher resolution. The technology of classification of bottom objects based on the training of an artificial neural network trained using synthesized training material is described. The results of checking the effectiveness of the developed algorithm using a digital polygon and a digital twin of AUV are presented. The simulation of the developed algorithm showed that the integrated use of bottom monitoring tools increases the likelihood of a successful solution to the problem and reduces the time needed to solve it.

• MODERN METHODS OF HYPERSPECTRAL IMAGE PROCESSING: SYSTEM ANALYSIS, ALGORITHMS AND PROSPECTS FOR APPLICATION IN CONSTRUCTION DIAGNOSTICS

  М. А. Filonova , S. N. Shirobokova

  188-208

  2026-07-07

  Abstract ▼

  The relevance of this study is determined by the growing interest in hyperspectral imaging as a tool for non-destructive testing of building materials and structures, as well as by the insufficient systematization of modern methods for processing such data. The aim of the work is to provide a systematic analysis of HSI processing algorithms, identify their advantages and limitations, and determine the prospects for their application in construction diagnostics. The study examines the specific features of hyperspectral data, including high dimensionality, noise, calibration errors, atmospheric distortions, and the shortage of labeled datasets. The evolution of approaches is shown: from classical machine learning methods and manual feature engineering to deep neural networks. Dimensionality reduction methods, kNN classifiers, Bayesian models, logistic regression, Random Forest, SVM, and MLP are analyzed, along with methods for incorporating spectral-spatial context. Special attention is paid to modern deep learning architectures: 1D, 2D, and 3D CNNs, RNNs, LSTM/GRU models, hybrid CNN–RNN models, transformers, and CNN–Transformer schemes. Transfer learning, semi-supervised learning, self-supervised learning, few-shot learning, meta-learning, and domain adaptation are considered separately as ways to overcome the limited availability of labeled data. The approaches are compared in terms of data requirements, computational complexity, robustness to noise, and their ability to account for spectral and spatial dependencies. It is shown that the most promising models for construction diagnostics are hybrid models that combine local convolutional features, the global context of attention mechanisms, and the possibility of fine-tuning on small specialized datasets. The paper summarizes the current state of the field and forms a basis for selecting methods for defect detection, moisture assessment, corrosion analysis, and evaluation of degradation in building structures. The conclusions formulated in the study can be used when designing experimental protocols and selecting architectures for further applied research in the field of building monitoring

• CLASSIFICATION FEATURES OF ENCRYPTED NETWORK TRAFFIC

  N. V. Boldyrikhin , D. A. Korochentsev , F.A. Altunin

  2020-10-11

  Abstract ▼

  Currently, there is growing interest in the tasks of efficient packet network management:
  quality of service, ensuring information security, optimization of the network hardware and software
  resources. All these tasks rely heavily on the analysis and classification of network traffic.
  This traffic is heterogeneous, as a rule, has a pulsating nature, difficult to predict and described by
  the mathematical apparatus of random processes. At different times, the conditions for passingpackets along the same path can vary significantly. At the same time, a significant number of applications
  are appearing requiring latency and jitter. The administration task in this context is to
  correctly configure the switching and routing nodes. Traffic classification allows you to identify
  packages of various applications and services and ensure their prioritization during transmission
  over the network. For example, video conferencing traffic needs to be transmitted first of all, since
  it is very sensitive to delays and jitter, data traffic can be transmitted last. The classification of
  traffic today is an urgent task both in terms of network administration and in terms of ensuring its
  security. Due to the fact that a large number of applications now encrypt the transmitted information
  and it is very difficult to view its contents, the traffic classification is of particular interest,
  which allows indirect signs to determine anomalies in the network, signs of intrusion. In this paper,
  we consider the features of solving the classification problem of encrypted traffic. The aim of
  the work is to study the classification features of encrypted traffic using correlation analysis and
  an algorithm based on the difference in integral areas. Research Objectives: – develop a traffic
  classification algorithm based on correlation and known patterns; – develop an algorithm based
  on the difference of the integral areas under the traffic intensity curves; – conduct a practical
  study of the accuracy of solving the classification problem. The work considers the classification
  of traffic into three groups: audio, video, data. As a result, a sufficient accuracy of the correlation
  algorithm in determining audio and data traffic was revealed. To identify video traffic, it is better
  to use an algorithm based on the difference of the integral areas under the intensity curves.