The article presents an approach to phishing attack detection based on the construction
and automatic optimization of machine learning pipelines using a specialized library (PhishAutoML). The urgency of the problem is due to the evolution of phishing attacks that use social
engineering methods and lexical tricks, which makes traditional static protection methods ineffective. The theoretical foundations of text vectorization using the FastText model and its application within an AutoML approach based on Bayesian optimization, which allows for the automatic selection of hyperparameters for the entire pipeline, are described. The proposed
PhishAutoML concept is used to build models capable of detecting phishing based on semantic
analysis and flexibly configuring the trade-off between quality and performance. The results of
computational experiments are presented: final metrics of quality and performance, as well as a
comparative analysis with classical (TF-IDF) and modern (DistilBERT) approaches. The conclusions confirm the effectiveness of the proposed solution (achieving a phishing detection recall of 95%, which is several times higher than alternative methods) and outline directions for
its further development. However, integrating AI into traditional intrusion detection and prevention systems poses significant risks and challenges. This article explores the key technical, organizational, and ethical barriers that hinder the widespread adoption of AI-powered solutions and
suggests potential solutions to overcome them.

This article presents a software package for automatic sentiment analysis of Uzbek texts. The system relies on a hybrid approach, combining a transformer model, a named entity extraction (NER) module, and a specially compiled sentiment dictionary of the Uzbek language. The relevance of this development is due to the growing volume of informal texts on social networks and the lack of ready-made tools for processing them. The package implements a full processing cycle: text cleaning and normalization, entity extraction, sentiment detection and keyword detection, and visualization of the results in a built-in web interface. The models and dictionary are adapted to the agglutinative and orthographic features of the Uzbek language, increasing resilience to colloquial and mixed forms of writing. The package's architecture, main software modules and their interactions, as well as the operating principle of the application interface (REST API) are briefly described. Examples of the system's application for analyzing user reviews and messages are provided, confirming its suitability for applied opinion monitoring tasks. Based on the results of initial experiments, a significant improvement in quality is achieved compared to basic text models without taking into account NER and lexicon.

 Currently, citizens actively participate in public discussions that affect the life of society. With the development of digitalization, users are increasingly using social networks and feedback platforms (FBPs) to express their opinions on various areas of activity. In this regard, there is a need to analyze public sentiment using modern analytical tools in order to identify
potential problems at an early stage. This article describes the development of a mathematical model for analyzing public sentiment based on text messages published on social networks and other sources. The model includes data preprocessing, clustering of requests, naming of clusters using language models, and summarization of identified problems.

The paper proposes a method for determining the coordinates of unmanned aerial vehicles (UAVs) based on measurements of ranges up to two ground stations that can be used in aviation navigation, radio positioning, and airspace surveillance. The method relies on an angular-difference rangefinder and forms the basis for a local navigation system used in difficult-toaccess areas. The goal of the study is to develop an approach for determining UAV positions based on measuring distances from the vehicle to a limited set of ground locations and estimating errors in these measurements, as well as establishing dependencies between these errors and
other geographical parameters related to the flight of the UAV. Mathematical modeling, numerical simulations, and experiments based on a developed simulation model are used as techniques. As a result of this work, dependencies of errors in determining the position of a UAV are obtained as a function of changes in key errors in input data and geometry of ground station locations. A new method for locating UAVs using ranges to two stations and barometric altimetry has been proposed. Errors associated with this method are also determined. It has been shown that accuracy of this method depends significantly on the geometry and accuracy of station locations, range measurements, and distance between stations and UAVs. The most favorable range for operation is between 30 and 150 degrees, where positioning errors remain minimal. Theoretical significance lies in a model for representing coordinates of UAVs, which forms the basis of the localization method. Practical significance is in presenting results from simulations of
method operation in various scenarios, as well as results from numerical experiments that demonstrate features of application and factors influencing measurement accuracy.

This paper provides an overview of the current state and prospects for applying fractal analysis to Earth remote sensing (ERS) data processing. It examines the evolution of the scientific paradigm–the transition from the concept of smooth, differentiable functions to a fractal paradigm that describes complex natural forms. The theoretical foundations of the fractal approach are analyzed in detail, including the concepts of fractal dimension, self-similarity, and anomalous stochastic processes. Particular attention is paid to the methodology for calculating the fractal dimension of natural objects using ERS data: the variogram method, the Box Counting algorithm, and triangulation methods. Applied aspects of fractal analysis for land cover classification, including polarimetric radar data and multispectral optical imagery, are also considered. The relationship between fractal characteristics, forest inventory parameters, and underlying surface structural features is analyzed. Further development prospects are discussed, including multifractal analysis and consideration of distorting factors during surveying