Search

Data streams can be defined as the continuous stream of data in many forms coming from different sources. Data streams are usually non-stationary with continually changing their underlying structure. Solving of predictive or classification tasks on such data must consider this aspect. Traditional machine learning models applied on the drifting data may become invalid in the case when a concept change appears. To tackle this problem, we must utilize special adaptive learning models, which utilize various tools able to reflect the drifting data. One of the most popular groups of such methods are adaptive ensembles. This paper describes the work focused on the design and implementation of a novel adaptive ensemble learning model, which is based on the construction of a robust ensemble consisting of a heterogeneous set of its members. We used k-NN, Naive Bayes and Hoeffding trees as base learners and implemented an update mechanism, which considers dynamic class-weighting and Q statistics diversity calculation to ensure the diversity of the ensemble. The model was experimentally evaluated on the streaming datasets, and the effects of the diversity calculation were analyzed.

This article presents a contribution to the general public education about the risks which people are facing on the web and how to prevent them. We first briefly characterize and present a typology of antisocial behavior on the web, analyze and compare some of the existing tools. Next, we present the design and implementation of a website to inform internet users about the possibilities of detecting antisocial behavior. A user study showed that the web application fulfills its purpose and is well accepted by the users.

This article about Active learning in data science education deals with methods and tools to support active learning of data science, in particular within the Knowledge Discovery subject at the Technical University of Kosice, Slovakia. One of the results of the work presented in this article is an implemented web application to support active learning in data science. The article involves a brief overview of the methods supporting the active learning which may be used also for teaching data science. The final web application which was proposed and tested by the users, was created by RShiny application. The supporting application is helpful mainly for the students as an interactive mean of active learning in data science. The result of the article is available and it was applied in practice. This work contributes to the effective learning of the users (learners) as well as it contributes to better understanding, comprehension and application of the acquired knowledge and experiences of the students in practice by means of active learning.

This paper approaches the problem of processing medical records with the aim of data analysis from a larger set of domain specific medical data. We solve this problem in cooperation with doctors from Department of Cardiology of the East Slovak Institute of Cardiovascular Disease. For this purpose, we designed, implemented and tested first prototype of a software system able to read and process medical records about cardiological patients and provide anonymized data for further analysis.

The heart disease describes a range of conditions affecting our heart. It can include blood vessel diseases such as coronary artery disease, heart rhythm problems or and heart defects. This term is often used for cardiovascular disease, i.e. narrowed or blocked blood vessels leading to a heart attack, chest pain or stroke. In our work, we analysed three available data sets: Heart Disease Database, South African Heart Disease and Z-Alizadeh Sani Dataset. For this purpose, we focused on two directions: a predictive analysis based on Decision Trees, Naive Bayes, Support Vector Machine and Neural Networks; descriptive analysis based on association and decision rules. Our results are plausible, in some cases comparable or better as in other related works.

Medical diagnostic is a complex process consisting of many input variables, which the general practitioner (GP) or specialist should take into account before confirm the expected diagnosis. In the case of electronic records, they have an opportunity to support this process within simple understandable results of the correctly applied suitable methods from machine learning or statistics. We used a small sample of patient's data from Croatia for experimental evaluation of this potential. We applied the methods as Welch's t-test, Pearson chi-square independence test, Youden's index, decision trees and simple K-Means. The cooperating medical expert evaluated the obtained results and confirmed the expected potential for daily medical practice.

Parkinson's disease is the second most frequent neurodegenerative disorder after Alzheimer's disease. There are numerous symptoms among the population suffering from the disease including tremor, slowed movement, impaired posture and balance, and rigid muscles, however dysphonia — changes in speech and articulation — is the most significant precursor. This is the reason why the article is focused on patients classification based on their speech signals. Algorithms C4.5, C5.0, RandomForest and CART were used to generate the decision trees in Rstudio interface. In addition, cut-off values of individual attributes were applied in order to classify the patients. The dataset in the article consists of 40 individuals' records, half of which were affected by Parkinson's disease. Each individual's coverage was represented by several records such as permanent vowels pronunciation, certain words, numbers and sentences. The objective was to determine the most accurate classification model (decision tree) using the individual types of speech signals. Cross-validation method was used for evaluation of models. The highest average model accuracy of 66.5% was obtained for data taken when individuals pronounced the numbers.

The aim of this article is to describe the design, implementation and evaluation of the educational application to support learning of data mining algorithms. The role of the application is to help students to better understand the algorithms such as Naive Bayes classifier, decision trees and association rules. The application also includes a test area that allows students to generate and solve different types of tasks on one hand side and teachers provide an effective way to test students without the need for creating custom tests on the other side. Presented application was evaluated from the perspective of students and teachers of the subject Knowledge Discovery, in order to verify the functionality and usability of the application in the real teaching process.

This article contains description how to create prediction models for more efficient utilization of running IT resources. Prediction model creation is presented as one of the most important steps within the process of the knowledge discovery. Presented models were created based on real data from a local IT company by means of the programming language R. This article also marginally covers the steps, which are to be carried out before the actual modelling. The steps to mention are the following: available data understanding and their subsequent preprocessing for further work, individual attributes analysis using data visualization by means of graphs and tracking certain behavior and correlations in data. First of all, designing models using linear and multiple regression will be followed in detail. Then, the methods of new values prediction will be described using the R language. Finally, the different types of prediction models will be considered using appropriate indicators. The results obtained should assist and simplify the methods of the knowledge discovery in certain enterprises based on the available databases, and also help to choose the most suitable prediction models learning method.

In this paper we present a proposal for a data-mining system deployed as a cloud service which is supposed to be used for a big data analysis. The main purpose of the system is the analysis of a vast number of event logs using means of data aggregation, clustering, classification and prediction. The system is composed of two components implemented as software services. The Automatic Predictor Generator is supposed to provide a meaningful way to aggregate large amounts of data and the Automatic Behavior Rule Extractor deals with proper analysis of these aggregations. Results of the system are the prediction rules usable for support of decision-making and in areas such as management, marketing, customer segmentation, classification, behavior prediction etc.

In this paper we describe the actual state of the big data analysis and related technologies teaching in Business information systems study program at the Technical University Kosice, Slovakia. We have been teaching at the Department of cybernetics and artificial intelligence, various courses specialized on data analysis, machine learning as well as knowledge discovery in databases. Advances and progress in technologies during recent years proved that there is a need to extend the teaching process by reflecting the new and emerging technologies, tools and approaches. During the last complex accreditation process we have proposed a new course specifically dedicated to the area of big data analysis and related technologies. The main objective of this paper is to summarize the motivation to include the newly introduced course within our curricula, as well as to describe the outline of the learning objectives, structure and content of the proposed course with focus on the design of the practical lab sessions.

This paper analyses how information from user profile influences quality of recommendations. We first start with an overview of recommendation systems, their functions methods used. The empirical part focuses on collaborative filtering method with the aim to find improvement of recommendations based on the user profile. The main objective for realized experiments was to verify the hypothesis that using information stored in user's profiles can improve collaborative filtering recommendation results. All our experiments were realized in RapidMiner tool on well-known MovieLens dataset. For evaluation of results we used standard metrics such as RMSE, MAE and NMAE. Experiments did not confirm the above mentioned hypothesis, but we present additional analysis of identified clusters with best and worst recommendation results.

Aspect-based sentiment analysis has become popular research field which allows the quantification of textual evaluations of different aspects of products and services. Methods of aspect-based sentiment analysis built on machine learning usually depend on manually annotated training corpora. In order to facilitate the processes of their creation, annotation tools dedicated to this purpose are needed. In this work we proposed a semi-automatic annotation tool which uses active learning to increase the effectiveness of the documents annotation. The use of active learning adapted to the needs of aspect-based sentiment analysis is the main difference between the proposed solution and existing annotation tools. We applied it in the domain of hotels evaluations. The results of realized experiments confirmed the faster increase of the annotation suggestions quality in terms of F1-measure in comparison to the scenario without active learning.

The paper addresses the concept of Technology Enhanced Learning, which is in focus of the IT4KT project of the Technical University of Kosice. Namely, we present an approach towards the collaborative design and implementation of semantic structures that are capable to support an automation in selected educational processes, especially in mathematical and computer science disciplines. The resulting structure of created ontologies is described together with generic process models that enable a semantic composition and orchestration of electronic services in the analyzed courses. Added semantics allows a flexible integration of educational services and processes, which consequently may help educators to minimize their efforts when preparing and running new instances of courses or lectures, as well as to effectively analyze the students' performance and the teaching process as whole.

E-learning represents still dynamically evolving area of research. For one of the sources of its persistent topicality can be considered its close relationship to the development in the fields of web services, semantic web, online collaboration, and many others, that is constantly reflected by e-learning solutions. The view of e-learning solutions, in which they were perceived as an independent solutions designed specifically to support educational processes, is changing, too. It is becoming increasingly common to integrate them into the wider context of Web portals of institutions and to make their content publicly accessible. A similar intent led us to design and implement the Web portal of Knowledge discovery course, which is taught at the Department of Cybernetics and Artificial Intelligence, Faculty of Electrical Engineering and Informatics at the Technical University of Kosice. Our intention was to create a solution to support educational processes, which would also represent a publicly available knowledge base dedicated to the topic of knowledge discovery, as part of the virtual laboratory of Business information systems. In this paper we describe the implementation details of the proposed solution built on the results of four ongoing projects. The solution is described both in terms of technological platform, based on the integration of existing open source solutions and our own extensions as well as in terms of the specifics of its content, which is based on video tutorials, collaboratively created documents and case studies.

This paper deals with social network analysis targeted to decomposition and local structure analysis of organization to organization network in order to get view of local structure characteristics in this network.

This paper proposes a client warehouse application over Microsoft Dynamics Navision 5.0 enterprise resource planning system. The application has been created to meet criteria and constraints stated by real world company, or in terms of application itself, the future user. By means of effectiveness, created application shortens or eliminates some warehouse activities. Employees can easily roam in warehouse and create physical inventory, transfer, intake or expend goods using PDAs with WiFi connection to database server. Furthermore, if WiFi signal is not strong enough to cover the whole working area of the warehouse, there is a possibility to synchronize essential data between source server database and local device compact database. Usage of barcode scanner makes every operation even more effective in terms of activity time consumption.

Sentiment analysis deals with methods for automatic analysis of the subjective aspects of the text. In this contribution we first present an overview of main approaches currently used in sentiment analysis. We further focus on feature selection methods for sentiment analysis and propose a new approach to feature selection. Our approach has been experimentally evaluated on movie review dataset. The results show that the proposed method is computationally efficient and in exchange sacrifices only a small amount of accuracy.

In this paper we analyze basic approaches for modeling of collaboration networks and propose some new extensions in weighting ties among event participants. One of the extensions is based on the idea of aging of the ties among collaborators. Classical as well as newly proposed approaches to weighting of ties in collaboration networks are experimentally evaluated. We performed two experiments with real data set and as a reference we used data gathered from peoples' opinions expressed in targeted inquiries.

This paper presents a short-term prediction of fog occurrence based on suitable data mining methods. The whole process was implemented through CRISP-DM methodology that represents most commonly used approach for data mining. This methodology consists of six main phases, which we describe in this paper for our application: business understanding, data understanding, data preparation, modeling, evaluation and deployment that resulted into new and useful knowledge to be used in real practice. The main motivation behind our solution was to develop an effective data mining model based on local conditions at the airport for short-term fog prediction as crucial factor for air management. Our first results presented in this paper are promising.

This paper describes design and implementation aspects for extension of our original software system developed in Java for support of information retrieval and text mining with specialized execution engine for different type of tasks. Some of our experiences and specific requirements of the real applications lead us to idea give the system possibility to run the tasks in distributive way. The result of the idea is task-based execution engine, which represents middleware-like transparent layer (mostly for programmers who want to re-use functionality of our package) for running of different tasks in multi-thread environment. The original system is being developed as open source with the intention to provide an easy extensible, modular framework for pre-processing, indexing and further exploration of large text collections. This specialized execution engine has been implemented within APVV project called PoZnaŤ (Support for knowledge creation processes) in order to extend proposed text-mining platform for educational and experimental purposes. Conceptual architecture of the system is provided as well as details regarding our extension of library like usage of content repository paradigm, representation and encapsulation of tasks, implementation of the engine itself and its role within PoZnaŤ platform.

Biomedical knowledge usually reaches the end users with a considerable lag behind the newest published discoveries. Even if we manage to collect texts of new experimental studies or clinical experiments on any particular topic, the data quantity usually exceeds the human capacity to process the data in a reasonable time. The problem is more prominent in the case of clinical genetics - sometimes we need an information on a heart defect for one patient, an hour later about a kidney malformation for another patient etc., while preparation of targeted recherche with the help of the most modern bibliographic tools takes from days to weeks. Thus, patients do not get a diagnostic care on the highest achievable level even in the most developed countries in the world. The major prominent genetic databases are e.g.: OMIM (Online Mendelian Inheritance in Man), MGI, GenBank, Entrez Nucleotide, Entrez Genome, Gene Ontology, Sanger Center, EOL, EnsEMBL. None of the named databases supports synergistic collaboration with any other named database. This article describes a system that improves an information retrieval among data provided by biomedical database NCBI (National center for biotechnology information).

This paper deals with the role of ontologies in collaborative systems. The research field of collaborative systems and their support by information and communication technologies describes theory called CSCW (computer supported cooperative work). During last few years the CSCW has been evolving and extending, e.g. CSCL (computer supported collaborative learning) has emerged. With a well-defined ontology structure, CSCL can accumulate the knowledge representation of learning objects, including participant background, group information, instruction designs, learning activities and learning outcomes.

The main goal of this work was analysis of excitation-emission fluorescent data. Data are obtained using luminescent spectrophotometer by measuring fluorescence intensity of specific samples. In particular, we had data obtained from samples containing substances, which are called fluorophores. Various fluorophores and their combinations were located at first in the ultra-clean water and next, in synthetic urine. The goal was to analyze the possibility of backward decomposition of complex measurements' data into single parts, corresponding to single fluorophores. As we show in this paper, this possibility has been experimentally verified as and brings a large potential for investigation of composite samples of various biological materials, which can lead e.g. to improved diagnostic methods in medicine.