Персона: Домашова, Дженни Владимировна
Email Address
Birth Date
Результаты поиска
Development of a generalized algorithm for identifying atypical bank transactions using machine learning methods
ИНФОРМАЦИОННО-АНАЛИТИЧЕСКАЯ СИСТЕМА РЕШЕНИЯ ЗАДАЧ МАШИННОГО ОБУЧЕНИЯ С УЧИТЕЛЕМ
Программа позволяет построить композицию произвольных моделей машинного обучения для определения классов заданных объектов. Программа поддерживает следующие этапы проведения анализа: предварительный анализ исходных данных пользователя, предобработку данных, отбор информативных признаков, построение композиций моделей машинного обучения, их обучение, представление характеристик результатов обучения с целью анализа полученного качества предсказания, а также применение обученных моделей для классификации. На каждом этапе пользователю предоставляется возможность выбора конкретных методов из доступного перечня алгоритмов обработки. Построенные и/или обученные композиции моделей могут быть сохранены для дальнейшего использования. ОС: Windows.
Constructor of compositions of machine learning models for solving classification problems
© 2020 The Authors. Published by Elsevier B.V.The article concerns the description of program software that gives an opportunity to solve classification problem by machine learning methods on input data in context of different subject areas. The list of such tasks can include: Identification of suspicious state contracts for collusion of suppliers, forecasting the execution of a government contract, forecasting license revocation from credit organizations and insurance companies, etc. The application allows a user to build a composition of several base machine learning models to solve classification problems. As part of the study, the functional requirements for the product being created are presented, the architecture has been developed, the design and testing of the proposed process for creating machine learning model compositions has been carried out. The use of the application will allow to build compositions of machine learning models in a user-friendly mode in order to increase the accuracy of the classification problem. The use of the proposed tools improves the accuracy of solving the classification problem on input data in the context of various subject areas through the utilization of a composition of machine learning methods.
Кластерный анализ стран на основе оценок ФАТФ
В данной статье проводится кластерный анализ стран на основе оценок взаимных оценок ФАТФ с использованием метода иерархической кластеризации (метод Уорда), алгоритма K-средних и графика локтя для определения оптимального числа кластеров. Применение этих методов позволило выявить устойчивые группы стран с схожими профилями соответствия стандартам ПОД/ФТ. Результаты исследования подтверждают возможность структурирования стран по оценкам ФАТФ, что открывает перспективы для более точной группировки стран по уровню соответствия ПОД/ФТ и повышения эффективности международного мониторинга.
Application of machine learning methods to identify suspicious actions of employees related to violation of the procedures of a credit institution
The article presents the results of the application of machine learning methods to identify suspicious actions of employees related to a violation of the procedures of a credit institution, specifically, the theft of funds from customer accounts and cards and abuse of the motivation system. The stages of data preprocessing within the considered task are analyzed. Among the considered classification algorithms, which are not sensitive to class imbalance, the method with the best value of hyperparameters was chosen. Next, the most informative features were highlighted, for which the best values of hyperparameters were selected and the optimal values of the probability thresholds of attributing an object to fraud were found. The proposed technology can be used separately or as part of an anti-fraud system for routine (for example, once a month) detection of illegitimate actions of employees of a credit institution related to the theft of funds from customer accounts and cards and abuse of the motivation system. A software tool in Python was developed that allows solving the task of detecting internal fraud based on the proposed technology.
Identification of atypical behavior of bank employees when using e-mail to prevent information leakages
The article presents the results of using machine learning methods to identify atypical behavior of bank employees when using e-mail. A feature space is formed that characterizes the behavior of e-mail users. The objects were previously clustered using the density-based spatial clustering of applications with noise (DBSCAN) and the fuzzy logic elements. The objects were marked using the inbuilt business rules, and the training sample was formed in the absence of marked data. The most informative features are selected, and a model of classification of e-mail users by the type of their behavior is constructed. A feature space is formed that defines the characteristics of a particular message to identify messages that are the information security incidents. Preliminary data processing was performed by removing the duplicates and encoding the categorical variables. A model of messages classification is constructed. The best combination of the machine learning method and the feature selection algorithm was determined using quality metrics. The constructed models allow specialists of cybersecurity departments of banks to identify employees with abnormal behavior and possibly involved in information leaks. A software tool in Python was developed that makes it easier to identify the final status of a message by partially replacing its manual detection for an automatic one.
Detection and analysis of atypical stock transactions with possible misuse of insider information and market manipulation: Methods and models
The article presents the results of the application of machine learning methods to identify customers prone to the use of insider information and market manipulation. The feature space was formed taking into account the specifics of the subject area. Cluster analysis techniques were used to markup data and to construct a training sample to identify features of anomalous objects. The key distinguishing features of atypical transactions were obtained and analyzed, an atypical investor profile was created. Ensemble algorithms of machine learning were used to build models for identifying transactions with possible misuse of insider information and signs of market manipulation. The practical significance of the study lies in the fact that the use of models will allow to effectively and timely identify atypical transactions, as well as unscrupulous investors involved in the misuse of insider information and market manipulation. The proposed approaches can be applied both in the work of the Bank of Russia and in the brokers work, who should independently monitor the doubtful operations of their clients.
Developing a named entity recognition model for text documents in Russian to detect personal data using machine learning methods
Selecting an optimal architecture of neural network using genetic algorithm
© 2020 Elsevier B.V.. All rights reserved.The article presents the results of applying a genetic algorithm to find the most optimal architecture of the neural network that would solve classification problem with minimal errors. The stages of the genetic algorithm are considered and the rule for encoding the parameters of the neural network is determined. A genetic algorithm for constructing the optimal architecture of a multilayer perceptron for solving classification problems has been developed. The algorithm independently creates a random population, evolves, creating new generations with more adapted individuals, i.e., neural networks with better architectures than previous generations. The paper describes the process of population formation, substantiates the choice of the fitness function and the method of selecting parents. Modifications of the crossover and mutation operators are proposed in order to ensure the operability of the algorithm on variable size individuals. A software tool that generates a neural network with the best parameters for solving classification problems has been developed on Python3 programming language. The architecture of a neural network for detecting fraudulent transactions has been built by using the developed software.