Machine Learning

Sleep is a basic human need, making up around one-third of our lives and being essential to the recovery of our physical well-being and general standard of living. However, poor sleep quality can interfere with these critical restorative processes, leading to disorders such as apnea and insomnia. These conditions not only impair daily performance but also have long-term health consequences. Furthermore, the challenges imposed by modern lifestyles have increased the prevalence of these sleep disorders, emphasizing the need for effective diagnostic tools. This research aims to harness the capabilities of Machine Learning (ML), specifically the Random Forest algorithm, to detect and analyze patterns indicative of sleep disorders in collected data sets. Random Forest is particularly suited for this task due to its ability to manage complex data sets by building multiple decision trees, thus creating a comprehensive and robust model for classifying sleep disorders. The findings of the study are promising, showing that the Random Forest algorithm can achieve a high level of accuracy in sleep disorder detection. The model demonstrated a test accuracy rate of 97.33%, with a precision of 96%, and a recall rate of 100%. Additionally, it achieved an F1-Score of 98% and a Kappa Score of 0.945, validating the reliability of this algorithm in producing precise classifications. This research offers significant insights into the patterns of sleep disorders and contributes to the development of targeted interventions aimed at improving sleep quality. Ultimately, this could significantly enhance the quality of life for individuals suffering from sleep disorders.

Multivariate measurements of human brain white matter (WM) with diffusion MRI (dMRI) provide information about the role of WM in a variety of cognitive functions and in brain health. Statistical models take advantage of the regularities in these data to make inferences about individual differences. For example, dMRI data provide the basis for accurate brain-age models -models that predict the chronological age of participants from WM tissue properties. Deep learning (DL) models are powerful machine learning models, which have been shown to provide benefits in many multivariate analysis settings. We investigated whether DL would provide substantial improvements for brain-age models based on dMRI measurements of WM in a large sample of children and adolescents. We found that some DL models fit the data better than a linear baseline, but the differences are small. In particular, recurrent neural network architectures provide up to ~6% improvement in accuracy. This suggests that information about WM development is mostly accessible with linear models, and does not require the additional invariance and non-linearity offered by DL models. However, in some applications this incremental improvement may prove critical. We provide open-source software that fits DL models to dMRI data ().

Multivariate measurements of human brain white matter (WM) with diffusion MRI (dMRI) provide information about the role of WM in a variety of cognitive functions and in brain health. Statistical models take advantage of the regularities in these data to make inferences about individual differences. For example, dMRI data provide the basis for accurate brain-age models – models that predict the chronological age of participants from WM tissue properties. Deep learning (DL) models are powerful machine learning models, which have been shown to provide benefits in many multivariate analysis settings. We investigated whether DL would provide substantial improvements for brain-age models based on dMRI measurements of WM in a large sample of children and adolescents. We found that some DL models fit the data better than a linear baseline, but the differences are small. In particular, recurrent neural network architectures provide up to ∼6% improvement in accuracy. This suggests that inform...

State of the art audio source separation models rely on supervised data-driven approaches, which can be expensive in terms of labeling resources. On the other hand, approaches for training these models without any direct supervision are typically high-demanding in terms of memory and time requirements, and remain impractical to be used at inference time. We aim to tackle these limitations by proposing a simple yet effective unsupervised separation algorithm, which operates directly on a latent representation of time-domain signals. Our algorithm relies on deep Bayesian priors in the form of pre-trained autoregressive networks to model the probability distributions of each source. We leverage the low cardinality of the discrete latent space, trained with a novel loss term imposing a precise arithmetic structure on it, to perform exact Bayesian inference without relying on an approximation strategy. We validate our approach on the Slakh dataset [1], demonstrating results in line with state of the art supervised approaches while requiring fewer resources with respect to other unsupervised methods.

Agriculture is crucial for the existence of humankind. Agriculture provides a significant portion of the income for many people all around the world. Additionally, it provides a large number of work possibilities for the general public. Numerous farmers desire for a return to the old-fashioned techniques of farming, which provides little profit in today’s market. Long-term economic growth and prosperity are dependent on the success of agriculture and associated companies in the United States. Agribusiness crop yields may be increased by carefully selecting the right crops and putting in place supportive infrastructure. Weather, soil fertility, water availability, water quality, crop pricing, and other factors are taken into consideration while making agricultural predictions. Machine learning is critical in crop production prediction because it can anticipate crop output based on factors such as location, meteorological conditions, and season. It is advantageous for policymakers and...

Data warehouses are environments used for data analysis and efficient decision making within companies. They are tools that allow the execution of complex and multidimensional queries. One of the security vulnerabilities that can be used by malicious users is data inference, which is the deduction of private information by devious means. In the present work, we tried to show that the existence of functional dependencies in the data can help to perform an inference attack by using supervised learning algorithms to infer private information. These algorithms are Support Vector Machine (SVM), Random Forest (RF), Bayesian Regularized Neural Network (BRNN) and K-Nearest Neighbors (K-NN). The BRNN provided a better performance in our study. This paper implements an inference attack using regression learning algorithms, studies different dependency situations in the data, and uses the combination of COUNT, SUM, AVG and STDEV queries. The use of several methods in this study allows the prevention of inferences when one of these methods is used by a malicious user. We managed to achieve this attack by detecting 09.12% inferences on all methods compared to BRNN whose realized inference rate is 03.94%.

One of the world wide health care concerns in the last two decades has been the decrease in fertility rates. The problem is said to be more severe among the male population. Research has shown that environmental factors and life style habits have an impact on the quality of semen. Orthodox diagnosis of seminal quality employs a laboratory approach, involving expensive tests, which are also sometimes uncomfortable to the patient. Application of machine learning techniques has been on the rise and has demonstrated encouraging results in many fields, including health care. In this paper we propose Naïve Bayes and Artificial Neural Network classifiers for the characterization of seminal quality, based on environmental factors and life style habits Comparisons between the two classifier models show that their accuracy rate is the same and stands at 97%, on the training set.

The unprecedented growth of computational capabilities in recent years has allowed Artificial Intelligence (AI) models to be developed for medical applications with remarkable results. However, a large number of Computer Aided Diagnosis (CAD) methods powered by AI have limited acceptance and adoption in the medical domain due to the typical blackbox nature of these AI models. Therefore, to facilitate the adoption of these AI models among the medical practitioners, the models' predictions must be explainable and interpretable. The emerging field of explainable AI (XAI) aims to justify the trustworthiness of these models' predictions. This work presents a systematic review of the literature reporting Alzheimer's disease (AD) detection using XAI that were communicated during the last decade. Research questions were carefully formulated to categorise AI models into different conceptual approaches (e.g., Post-hoc, Ante-hoc, Model-Agnostic, Model-Specific, Global, Local etc.) ...

To be a genuinely useful methodology for cognitive psychology, situated action must be extended to explain higher-level human cognition. Yet there is a tension between the requirements for cognitive explanations and the kind of explanations situated action supplies. It is just not clear how to use embodied interaction with an external environment to explain processes that are so internal and that seem to crucially involve representatious and concepts. In short, there is nothing for concepts to interact with. Perhaps they interact with each other. This is the idea we explore here. We give a situated action style explanation for conceptual change in the context of analogical reminding. We have two central results: 1) analogical reminding changes concepts in long-term memory, and 2) concepts have implicit as well as explicit structure.

The COVID-19 pandemic has precipitated an unprecedented surge in the proliferation of online E-learning platforms, designed to cater to a wide array of subjects across all age groups. However, a paucity of these platforms adopts a learner-centric approach or validates user learning, underscoring the need for effective E-learning validation and personalized learning recommendations. This paper addresses these challenges by implementing an innovative approach that leverages real-time electroencephalogram (EEG) signals collected from learners, who don neuro headsets while partaking in online courses. These EEG signals are subsequently classified using Convolutional Neural Networks (CNN) and Long Short-Term Memory (LSTM) deep learning models, with the intent of discerning the efficacy of the E-learning process. The proposed models have yielded promising classification accuracies of 68% and 97% for the CNN and LSTM models, respectively, demonstrating their rapidity and precision in classifying E-learning EEG signals. Thus, these models hold substantial potential for application in similar E-learning validation scenarios. Furthermore, this study introduces an automated framework designed to track the learning curve of users and furnish valuable recommendations for E-learning materials. The presented approach, therefore, not only validates the E-learning process but also aids in optimizing the learning experiences on E-learning platforms.

Research has shown that lexical databases are good sources of lexical knowledge for various Natural Language Processing (NLP) tasks. Over the years, several lexical databases have been developed, eg HowNet [1], WordNet [2], EuroWordNet [3] and CCD [4]. These knowledge ...

In supervised learning the inductive algorithm seeks to develop a conceptual description, or prescriptive model, from examples or objects that have been pre-classified. On the other hand, in unsupervised learning, or clustering, the task of the algorithm is to group non-classified examples into clusters of examples, building a descriptive model. Central to all of the goals of clustering is the notion of similarity (or dissimilarity) between the individual objects being clustering. The key distinction between prediction and description is that prediction has as its objective a unique variable, known as the class attribute in supervised learning, while in descriptive problems no single variable is central to the problem. In this work we propose an approach, as well as a particular implementation of this approach, were a combination of traditional clustering and inductive learning are used to find and interpret special patterns from a database in order to extract useful knowledge to represent real world domains. This approach helps the user to make explicit his/her prior knowledge relevant to the problem allowing to organize the set of objects in different clusters into a set of classes considered to be semantically interpretable by the user.

In supervised learning the inductive algorithm seeks to develop a conceptual description, or prescriptive model, from examples or objects that have been pre-classified. On the other hand, in unsupervised learning, or clustering, the task of the algorithm is to group non-classified examples into clusters of examples, building a descriptive model. Central to all of the goals of clustering is the notion of similarity (or dissimilarity) between the individual objects being clustering. The key distinction between prediction and description is that prediction has as its objective a unique variable, known as the class attribute in supervised learning, while in descriptive problems no single variable is central to the problem. In this work we propose an approach, as well as a particular implementation of this approach, were a combination of traditional clustering and inductive learning are used to find and interpret special patterns from a database in order to extract useful knowledge to re...

In this paper we explore the effects of query and database size on news story classification performance. Memory Based Reasoning (MBR) (a k-nearest neighbor method) used as the classification method. There are 360 different possible codes. Close matches to a new story are found using an already coded training database of about 87,000 stories from the Dow Jones Press Release News Wire, and a Connection-Machine Document Retrieval system (CMDRS, [Stanfill]) that supports full text queries, as the underlying match engine. By combining the codes from the near matches, new stories are coded with a recall of about 80% and precision of about 70%, as reported in [Masand]. When the query size is varied from 10 terms to more than 200 terms (matched against the full-text of documents) the recallprecision product changes from 0.53 to 0.6. While this is a significant change, we find that moderate sized queries of 40-80 terms can suffice for finding relevant matches for classification. By changing the size of database from 10,000 stories to 80,000 we found that the recall-precision product changed from 0.22 to 0.57. This shows that with our current MBR approach the database size can't be reduced significantly without compromising performance. We also find that fewer number of retrieved matches are needed with larger query and database sizes.

Next-generation transportation surveys will utilize Global Positioning Systems (GPS) to collect trip data. Due to their ubiquity, GPS-enabled mobile devices are becoming promising for use as survey tools. TRAC-IT is a mobile phone application that records travel behavior by collecting real-time GPS data and requiring minimal input from the user for data such as trip purpose, mode of transportation, and vehicle occupancy. To ease survey burden on participants, new techniques must be explored to derive more information directly from GPS data. This paper demonstrates the feasibility of using neural networks and assisted GPS data collected from GPS-enabled mobile phones to automatically detect the mode of transportation. Furthermore, this paper demonstrates that this technique can be optimized using a critical point algorithm to reduce the size of required GPS datasets obtained from GPS-enabled mobile phones, thus reducing data collection costs while saving mobile phone resources such as battery life.

SUMMARY In this paper the LUSAS High Precision Moulding (HPM) integrated finite element environment for composite process simulation is introduced and validated against test components with single and double curvatures. Presented results show dependence on the laminate thickness, radius and in-plane dimensions, these agree well with the experimental data.

Smoking cessation is an important public health policy worldwide. However, as far as we know, there is a lack of screening of variables related to the success of therapeutic intervention (STI) in Brazilian smokers by machine learning (ML) algorithms. To address this gap in the literature, we evaluated the ability of eight ML algorithms to correctly predict the STI in Brazilian smokers who were treated at a smoking cessation program in Brazil between 2006 and 2017. The dataset was composed of 12 variables and the efficacies of the algorithms were measured by accuracy, sensitivity, specificity, positive predictive value (PPV) and area under the receiver operating characteristic curve. We plotted a decision tree flowchart and also measured the odds ratio (OR) between each independent variable and the outcome, and the importance of the variable for the best model based on PPV. The mean global values for the metrics described above were, respectively, 0.675±0.028, 0.803±0.078, 0.485 ±0.146, 0.705±0.035 and 0.680±0.033. Supporting vector machines performed the best algorithm with a PPV of 0.726±0.031. Smoking cessation drug use was the roof of decision tree with OR of 4.42 and importance of variable of 100.00. Increase in the number of relapses also promoted a positive outcome, while higher consumption of cigarettes resulted in the opposite. In summary, the best model predicted 72.6% of positive outcomes correctly. Smoking cessation drug use and higher number of relapses contributed to quit smoking, while higher consumption of cigarettes showed the opposite effect. There are important

In software development systems, the maintenance process of software systems attracted the attention of researchers due to its importance in fixing the defects discovered in the software testing by using bug reports (BRs) which include detailed information like description, status, reporter, assignee, priority, and severity of the bug and other information. The main problem in this process is how to analyze these BRs to discover all defects in the system, which is a tedious and time-consuming task if done manually because the number of BRs increases dramatically. Thus, the automated solution is the best. Most of the current research focuses on automating this process from different aspects, such as detecting the severity or priority of the bug. However, they did not consider the nature of the bug, which is a multi-class classification problem. This paper solves this problem by proposing a new prediction model to analyze BRs and predict the nature of the bug. The proposed model constructs an ensemble machine learning algorithm using natural language processing (NLP) and machine learning techniques. We simulate the proposed model by using a publicly available dataset for two online software bug repositories (Mozilla and Eclipse), which includes six classes: Program Anomaly, GUI, Network or Security, Configuration, Performance, and Test-Code. The simulation results show that the proposed model can achieve better accuracy than most existing models, namely, 90.42% without text augmentation and 96.72% with text augmentation. INDEX TERMS Software maintenance, nature classification, ensemble machine learning algorithm, natural language processing, bug reports, machine learning.

STRIKE is a diagnostic calorimeter composed of 16 CFC tiles with unidirectional properties used to study the beams of particles generated in the SPIDER experiment. Two thermal cameras will be used to analyze the temperature of the tiles and reconstruct the bidimensional flux of energy striking the calorimeter. Most of the conventional methods used to evaluate the inverse heat flux are unbearably time consuming; since the objective is having a tool for heat flux evaluation for STRIKE real time operation, the need to have a ready-to-go instrument to understand the beam condition becomes stringent. For this reason a neural network was chosen to perform this analysis. During the thesis work an existing convolutional neural network was optimized to retrieve the parameters of the beam from the thermographic images. In particular, starting from a network able to determine the position and the radius of a singular circular shape on a noiseless background, the possibility to recognize the po...

В данной статье рассматривается применение различных моделей к решению задачи детекции товаров различных брендов на изображении. Проведен сравнительный анализ наиболее эффективных и широко применяемых нейросетевых архитектур Faster R-CNN (ResNet50/101), SSD и YOLO. Полученные результаты подтверждают эффективность применения архитектуры Faster R-CNN к любым выборкам изображений. Однако нужно отметить ресурсоемкость таких архитектур и их непригодность для решения задач, в котором важным критерием эффективности является время выполнения анализа. Модели SSD и YOLO, хотя и не дают преимуществ при детекции объектов малых и средних размеров, могут успешно применяться в составе мобильных систем детекции, ограниченных в своих аппаратных возможностях. Кроме этого, данные нейросетевые архитектуры выполняют обработку быстрее Faster R-CNN и могут рассматриваться как базовые модели для детекции и сегментации фото- и видеоизображений в реальном времени

В данной статье рассматривается и анализируется применение метода word2vec для решения задач семантического кодирования. На основании полученных результатов разработан прототип поисковой системы, базирующейся на использовании выделенной семантической информации для осуществления релевантного поиска в базе документов. Предложено два основных сценария осуществления такого поиска. Осуществлена подготовка обучающей выборки на базе корпуса документов англоязычной версии Википедии, включающей более 100 тысяч оригинальных статей. Полученная выборка использовалась в экспериментальной части работы для проверки эффективности разработанного прототипа поисковой системы

VAITSEKHOVICH L. U., GOLOVKO V. A., KUROSH MADANI Construction of system of detection of attacks with use the column of interaction of the agentsВ работе представлена мультиагентная модель системы обнаружения сетевых атак. Описана ее структура и рассмотрен алгоритм функционирования. В данной модели используется многослойная архитектура иерархии агентов, при чем алгоритм их взаимодействия описан набором правил, представленных в виде направленного графа. Модель может выполнять классификацию сетевых атак по классам и типам активности. Эксперименты свидетельствуют о том, что такая модель способна сократить число ложных срабатываний системы. Для обучения нейросетевых агентов использованы сгенерированные образы псевдоатак

Из-за постоянно возрастающего числа удалённых атак с использованием протоколов стека TCP/IP технологии обнаружения атак имеют большое значение для защиты компьютерных сетей. Существует множество систем для обнаружения атак, которые основываются на различном применении технологий определения аномального поведения. Данная статья описывает применение нейронных сетей для обнаружения и распознавания атак. Она базируется на использовании нелинейной рециркуляционной нейронной сети и многослойного персептрона. Для обучения и тестирования нейронных сетей применяется база данных KDD. Обсуждаются результаты экспериментов

VOITSCEKHOVICH L.Y., GOLOVKO V.A., KOCHURKO P.A., VOITSCEKHOVICH G.Y. Intrusion detection system as a basic protection element of computer networksВысочайший уровень угроз информационной безопасности из внешней среды сделал брандмауэр и Систему Обнаружения Вторжений необходимой составляющей защищенной информационной системы. В статье приводится классификация систем обнаружения вторжений, их структура, рассматриваются основные аспекты построения и развертывания в сети, а также представлены наши разработки в этой области. Цель статьи – обосновать необходимость проведения дальнейших исследований и предложить новый подход к построению систем обнаружения вторжений

KOCHURKO P. A., GOLOVKO V. А. Construction neuronetwork of system of detection and recognition of attacksРассматриваются вопросы построения нейросетевой системы для решения задач обнаружения и распознавания типов атак на базе совокупного классификатора из частных нейросетевых детекторов, отличающейся способностью к самоорганизации и обнаружению неизвестных атак. Описывается модульная структура системы, приводятся алгоритмы функционирования и способы реализации. Производится экспериментальное тестирование построенного макета системы в реальном времени

GOLOVKO V.A., BEZOBRAZOV S.V., MELENCHUK V.V. Neural network approach for spam detectionРассмотрен нейросетевой подход для обнаружения электронного спама. Рассмотрены основные методы и алгоритмы, а также их недостатки, применяемые для обнаружения нежелательной электронной корреспонденции. Разработана система обнаружения спама на основе применения искусственных нейронных сетей. Представлены результаты исследований

GOLOVKO V. A., MIKHNO E. V., BRICH A. L., MIKHNYAEV A. L., VOYTSEKHOVICH L. Yu., MATYUSHKOV A. L. The reduced svertochny neural network for exact recognition of hand-written figuresВ данной статье рассматривается сверточная нейронная сеть для распознавания рукописных цифр. Разработана упрощенная архитектура сверточной нейронной сети, которая позволяет классифицировать рукописные цифры с точностью 99.29% (ошибка тестирования 0.71%), что является лучшим результатом в классе неглубоких (shallow) сверточных нейронных сетей

This paper presents and explains an implementation of Learning Vector Quantization neural network for tennis activity detection and recognition using PIQ ROBOT device. The gesture recognition market is estimated to grow from 2013 till 2018 and is expected to cross $15.02 billion by the end of these five years. Analysts forecast the Global Gesture Recognition market to grow of 29.2 percent over the period 2013-2018. In terms of industry it means that currently consumer electronics application contributes to more than 99 percents of the global gesture recognition market. This paper starts our ambitious research in the area of artificial neural networks for activity recognition.

The wealth of smartphone data collected by the Cosmic Ray Extremely Distributed Observatory (CREDO) greatly surpasses the capabilities of manual analysis. So, efficient means of rejecting the non-cosmic-ray noise and identification of signals attributable to extensive air showers are necessary. To address these problems we discuss a Convolutional Neural Network-based method of artefact rejection and complementary method of particle identification based on common statistical classifiers as well as their ensemble extensions. These approaches are based on supervised learning, so we need to provide a representative subset of the CREDO dataset for training and validation. According to this approach over 2300 images were chosen and manually labeled by 5 judges. The images were split into spot, track, worm (collectively named signals) and artefact classes. Then the preprocessing consisting of luminance summation of RGB channels (grayscaling) and background removal by adaptive thresholding was performed. For purposes of artefact rejection the binary CNN-based classifier was proposed which is able to distinguish between artefacts and signals. The classifier was fed with input data in the form of Daubechies wavelet transformed images. In the case of cosmic ray signal classification, the well-known feature-based classifiers were considered. As feature descriptors, we used Zernike moments with additional feature related to total image luminance. For the problem of artefact rejection, we obtained an accuracy of 99%. For the 4-class signal classification, the best performing classifiers achieved a recognition rate of 88%.

The Vertically Integrated Project (VIP) model, developed in engineering, is a methodological approach that integrates undergraduate students, graduate students and faculty in projects for both research and student training. Recent applications of this model report that it could foster student-faculty collaboration, empower senior students to mentor juniors, transcend disciplinary silos that often hinder research, and promote diversity and inclusivity in higher education. Our cross-disciplinary experimentation with this model in ethnographic methods training demonstrates its value to facilitate or even institutionalize faculty-student collaborative research and mutual learning in social sciences. We also suggest transforming this model to be less hierarchical in collaboration, more mutually beneficial for all involved, and more sustainable for long-term projects.

A common practice in pre-processing step of hydrological modeling is to ignore observations with any missing variable values at any given time step even if it is only one of the independent variable that is missing. These rows of data are labeled incomplete and would not be used in either model building or subsequent testing and verification steps. This is not necessarily the best way of doing it as information is lost when incomplete rows of data are thrown out. Learning algorithms are affected by such problems more than physically-based models as they rely heavily on the data to learn the underlying input/output relationships. In this study, the extent of damage to the performance of the learning algorithm due to missing data is explored in a field-scale application. We have tested and compared the performance of two well- known learning algorithms, namely Artificial Neural Networks (ANNs) and Support Vector Machines (SVMs) for short-term prediction of groundwater levels in a well...

Background Prediction of short-term outcomes in young patients with acute ischemic stroke (AIS) may assist in making therapy decisions. Machine learning (ML) is increasingly used in healthcare due to its high accuracy. This study aims to use a ML-based predictive model for poor 3-month functional outcomes in young AIS patients and to compare the predictive performance of ML models with the logistic regression model. Methods We enrolled AIS patients aged between 18 and 50 years from the Third Chinese National Stroke Registry (CNSR-III), collected between 2015 and 2018. A modified Rankin Scale (mRS) ≥ 3 was a poor functional outcome at 3 months. Four ML tree models were developed: The extreme Gradient Boosting (XGBoost), Light Gradient Boosted Machine (lightGBM), Random Forest (RF), and The Gradient Boosting Decision Trees (GBDT), compared with logistic regression. We assess the model performance based on both discrimination and calibration. Results A total of 2268 young patients with a mean age of 44.3 ± 5.5 years were included. Among them, (9%) had poor functional outcomes. The mRS at admission, living alone conditions, and high National Institutes of Health Stroke Scale (NIHSS) at discharge remained independent predictors of poor 3-month outcomes. The best AUC in the test group was XGBoost (AUC = 0.801), followed by GBDT, RF, and lightGBM (AUCs of 0.795, 0, 794, and 0.792, respectively). The XGBoost, RF, and lightGBM models were significantly better than logistic regression (P < 0.05). Conclusions ML outperformed logistic regression, where XGBoost the boost was the best model for predicting poor functional outcomes in young AIS patients. It is important to consider living alone conditions with high severity scores to improve stroke prognosis.

A 2021 második felétől kibontakozó energiaár-robbanás, amelyet a lakosságot kevésbé, inkább a közszféra fogyasztóit súlytotta, olyan költségtakarékos megoldásokat követeltek meg a piaci szereplőktől, amelyek jellemzően jelentős beruházást igényelnek és hosszú távon térülnek csak meg. Jelen cikkben a 2020-2022 időszakra fókuszáltan, a témában releváns szakirodalmi forrásokon alapuló adatgyűjtést végeztem, amely az elemzéssel együtt a konstruktivista módon megalapozott kutatási stratégiák mentén zajlott. A két vizsgált kutatási kérdés arra kereste a választ, hogy a magyar szállodapiac szereplői 2020-2022-ben milyen kihívásokkal szembesültek, és milyen tényezők befolyásolják negatívan üzleti tevékenységüket, illetve, hogyan reagáltak a szállodaüzemeltetők az azonosított kihívásokra, és ezek a válaszok hogyan illeszkednek a globális legjobb gyakorlatokhoz. A magyar szállodák általános költségtakarékossági gyakorlatának vizsgálata mellett olyan megoldásokra helyeződik a hangsúly, amelyek a vendégélményt növelik, és a költséghatékony üzemeltetés terén valódi megoldást jelentenek az üzemeltetőktől. Az értékteremtés a szállodai működés különböző területein valósul meg, lefedve a front és back munkaterületeket egyaránt. Olyan megoldások kerülnek bemutatásra, mint a fűtési rendszer korszerűsítése, a digitalizáció, a szervezetfejlesztés vagy a tevékenységek kiszervezése.

A large literature studies the predictability of stock returns by other lagged financial variables in a predictive regression setting. A common feature of widely used testing procedures is a failing robustness, which may lead to misleading conclusions determined by the particular features of a small subfraction of the data. We propose a new general method to deal with this problem based on the robust subsampling approach. The method implies robust confidence intervals and inference results. It is applicable both in the multipredictor context and in settings with nearly integrated regressors. Simulation evidence confirms the higher accuracy and efficiency of our robust testing approach for typical applications in which the data may follow only approximately a predictive regression model. We apply our approach to US equity data from 1961 to 2008 and find that it yields a stronger evidence in favor of predictability than a number of other (nonrobust) tests in the literature.

The majority of machine learning methods and algorithms give high priority to prediction performance which may not always correspond to the priority of the users. In many cases, practitioners and researchers in different fields, going from engineering to genetics, require interpretability and replicability of the results especially in settings where, for example, not all attributes may be available to them. As a consequence, there is the need to make the outputs of machine learning algorithms more interpretable and to deliver a library of "equivalent" learners (in terms of prediction performance) that users can select based on attribute availability in order to test and/or make use of these learners for predictive/diagnostic purposes. To address these needs, we propose to study a procedure that combines screening and wrapper approaches which, based on a user-specified learning method, greedily explores the attribute space to find a library of sparse learners with consequent low data collection and storage costs. This new method (i) delivers a low-dimensional network of attributes that can be easily interpreted and (ii) increases the potential replicability of results based on the diversity of attribute combinations defining strong learners with equivalent predictive power. We call this algorithm "Sparse Wrapper AlGorithm" (SWAG).

Design Smell detection has proven to be an effective strategy to improve software quality and consequently decrease maintainability expenses. In this work, we explore the influence of the size category of the software project on the automatic detection of God Class Design Smell by different machine learning techniques. A set of experiments were conducted with eight different learning classifiers on a dataset formed by 12,588 classes of 24 systems. The results were evaluated using ROC area and Kappa tests. The classifiers change their behaviour when they are used in sets that differ in the value of the selected size information of their classes. This study concludes that it is possible to improve results, mainly in agreement, of God Class detection feeding machine learning classifiers with project size information of the classes to analyze.

Regression is a very relevant problem in machine learning, with many different available approaches. The current work presents a comparison of a large collection composed by 77 popular regression models which belong to 19 families: linear and generalized linear models, generalized additive models, least squares, projection methods, LASSO and ridge regression, Bayesian models, Gaussian processes, quantile regression, nearest neighbors, regression trees and rules, random forests, bagging and boosting, neural networks, deep learning and support vector regression. These methods are evaluated using all the regression datasets of the UCI machine learning repository (83 datasets), with some exceptions due to technical reasons. The experimental work identifies several outstanding regression models: the M5 rule-based model with corrections based on nearest neighbors (cubist), the gradient

As a result of social network popularity, in recent years, hate speech phenomenon has significantly increased. Due to its harmful effect on minority groups as well as on large communities, there is a pressing need for hate speech detection and filtering. However, automatic approaches shall not jeopardize free speech, so they shall accompany their decisions with explanations and assessment of uncertainty. Thus, there is a need for predictive machine learning models that not only detect hate speech but also help users understand when texts cross the line and become unacceptable. The reliability of predictions is usually not addressed in text classification. We fill this gap by proposing the adaptation of deep neural networks that can efficiently estimate prediction uncertainty. To reliably detect hate speech, we use Monte Carlo dropout regularization, which mimics Bayesian inference within neural networks. We evaluate our approach using different text embedding methods. We visualize the reliability of results with a novel technique that aids in understanding the classification reliability and errors.

The use of information technology using the internet is very easy to find information, so its users do not have to go to the place to find information, website and mobile applications internet users can directly get the information you want. Every Manager world tourism now provides the details about the tourism products offered. Very useful information at this time because people tend to look for quick information in the booking through the review of others in social media, blogs and websites. The importance of the review of the hotel as a source of information for travelers will plan trips. Currently known methods of classification of the highest accuracy in classifying hotels Indonesia-language review. So I need to know how Naive Bayes algorithms of accuracy, Support Vector Machine, Decission Tree (C4.5) and Naive Bayes Method with Particle Swarm Optimization Feature Selection. The results obtained from the comparison of four methods of such algorithms, a better level of accuracy ...

Group recommendation systems can be very challenging when the datasets are sparse and there are not many available ratings for items. In this paper, by enhancing basic memorybased techniques we resolve the data sparsity problem for users in the group. The results have shown that by conducting our techniques for the users in the group we have a higher group satisfaction and lower group dissatisfaction.

The recognition of unseen objects from a semantic representation or textual description, usually denoted as zeroshot learning, is more prone to be used in real-world scenarios when compared to traditional object recognition. Nevertheless, no work has evaluated the feasibility of deploying zero-shot learning approaches in these scenarios, particularly when using low-power devices. In this paper, we provide the first benchmark on the inference time of zero-shot learning, comprising an evaluation of state-of-the-art approaches regarding their speed/accuracy trade-off. An analysis to the processing time of the different phases of the ZSL inference stage reveals that visual feature extraction is the major bottleneck in this paradigm, but, we show that lightweight networks can dramatically reduce the overall inference time without reducing the accuracy obtained by the de facto ResNet101 architecture. Also, this benchmark evaluates how different ZSL approaches perform in low-power devices, and how the visual feature extraction phase could be optimized in this hardware. To foster the research and deployment of ZSL systems capable of operating in real-world scenarios, we release the evaluation framework used in this benchmark ().