We have developed a method called allocation, which is intended to classify imbalanced data. The allocation method is a classification ensemble composed of two levels. In the first level, there is the allocator, an algorithm of unsupervised learning, which learns to split the original dataset to homogeneous subsets and allocates them to specialized classifiers on the second level. The second level consists of multiple specialized classifiers where each learns on the specific subset of instances allocated to it and so specializes in a particular type of data. These specialized classifiers return the class of the instances. To test the concept of the allocation method, we developed the allocator with anomaly detection, which uses the one-class SVM classifier. The allocator was tested in combination with the six basic classification methods as the specialized classifiers on the second level. All variants and the combinations of the allocation method were tested on unbalanced and balanced datasets, the latter for the purpose of validation of the allocation as a general classification approach. The results of the allocation method were compared with existing methods for dealing with unbalanced data. The results of the experiments were further analysed with statistical methods, with which we confirmed that the allocation method is an effective alternative to the existing approaches for the classification of unbalanced and balanced data. Based on the results we did a follow up experiment where we tested the allocation method on unbalanced data thoroughly and in great detail. The results were published in a paper [COBISS.SI-ID 20206870].
COBISS.SI-ID: 19263510
The number of design patterns has been rapidly rising, but management and searching facilities appear to be lagging behind. This is why we developed a novel approach, which is used to search for suitable design patterns. It is built in an intelligent component on top of our platform for design patterns knowledge exchange. The design and development of the platform, as well as the approach validation was done in separate research, which was also published (A question-based design pattern advisement approach. [COBISS.SI-ID 18211350]). In our approach the developer no longer searches for an appropriate design pattern, but rather the intelligent component asks the developer questions. Our novel approach does not need extra effort in terms of maintaining a special expert system. Guided dialogues consist of independent questions from different sources and authors that are automatically combined. The approach is enabled with novel mathematical model and algorithms. They are based on continuous entropy-base measurement of gathered knowledge on design problem and maximizing its value. Hypothesis (our approach advantage in terms of correctness level at considerable lower input effort) validation was performed via comparison with human-created expert systems with a decision tree. Experiments were executed in order to verify our approach performance. The control group used a human-created expert system, while others were given a proposing component.
COBISS.SI-ID: 20216342
In advanced heterogeneous systems, which store their data primarily in local relational databases, using different formats and various local concepts, harmonising the metadata format alone does not solve the issue of efficient access to relevant information in heterogeneous environments, when different systems use different content, contextual and semantic concepts for certain entities. In this research, we have developed and proposed a new ontologically supported semantic search engine (OSSSE) which, in addition to the harmonisation of the metadata format among local information systems, also ensures that the meaning of data and/or concepts that belong to various metadata entities are also harmonised. A special model of ontological infrastructure was designed, and dedicated test ontology was created alongside with a new simplified algorithm for creating ontology, the basis of which is the distinction between new and already existing classes in terms of content. We evaluated the proposed OSSSE model using a simulation of the search process on the real searches within SICRIS. The obtained results show that regardless of the search situation, the proposed OSSSE is always at least as efficient as a search without ontological support in terms of precision, while recall remains the same; the improvement has been shown to be statistically significant with a high confidence interval (p(0.005). The proposed OSSSE model is able to solve the issue of harmonizing the data where different heterogeneous systems use different contents, contextual and semantic concepts, which is the case in many advanced expert systems. The technological basis for mapping ontologies to objects using a transformation based on description logics, which can be used for straightforward development of semantic expert systems, has been presented in [COBISS.SI-ID 18140950].
COBISS.SI-ID: 19798806
The increasing availability of electronic health care records has provided remarkable progress in the field of population health. The identification of disease risk factors has flourished under the surge of available data. Researchers can now access patient data across a broad range of demographics and geographic locations. Utilizing this Big healthcare data researcher have been able to empirically identify specific high-risk conditions found within differing populations. However, to date most studies approached the issue from the top down, focusing on the prevalence of specific diseases within a population. Through our research, we demonstrate the power of addressing this issue bottom-up by identifying specifically which diseases are higher-risk for a specific population. We demonstrate that network-based analysis can present a foundation to identify pairs of diagnoses that differentiate across population segments. Some of the concepts that were used to develop this approach have their roots in our research work together with Stanford University that was published in the high impact factor journal in 2015 [COBISS.SI-ID: 2145444]. We provide a case study highlighting differences between high and low income individuals in the United States. This work is particularly valuable when addressing population health management within resource-constrained environments such as community health programs where it can be used to provide insight and resource planning into targeted care for the population served. Some outcomes of this work were further developed in our SCOPES project.
COBISS.SI-ID: 2236324
In recent years, social networking sites have spread rapidly, raising new issues in terms of privacy and self-disclosure online. For a better understanding of how privacy issues determine self-disclosure, a model which includes privacy awareness, privacy social norms, privacy policy, privacy control, privacy value, privacy concerns and self-disclosure was built. A total of 661 respondents participated in an online survey and a structural equation modelling was used to evaluate the model. The findings indicated a significant relationship between privacy value/privacy concerns and self-disclosure, privacy awareness and privacy concerns/self-disclosure, privacy social norms and privacy value/self-disclosure, privacy policy and privacy value/privacy concerns/self-disclosure, privacy control and privacy value/privacy concerns. The model from the study should contribute new knowledge concerning privacy issues and their shaping of self-disclosure on social networking sites. It could also help networking sites service providers understand how to encourage users to disclose more information. Based on this study we also published other contributions at scientific conferences - [COBISS.SI-ID 15827990], [COBISS.SI-ID 15052566] and [COBISS.SI-ID 15499798].
COBISS.SI-ID: 18341142
Successful modelling tools need to effectively support individual as well as team-based work (collaboration) within collocated and virtual environments. In the past, achieving this has been challenging, since traditional modelling tools are desktop-based and thus suitable for individual and collocated work only. However, with the rise of web-based architectures and the cloud paradigm, desktop modelling tools now have rivals in their web-based counterparts that are especially suited for online collaboration (e-collaboration). The objective of our research was to probe the question as to 'which type of modelling tools (desktop or cloud-based) performs better in cases of individual work and e-collaboration', and to obtain insights into the sources of the strengths and weaknesses regarding both types of modelling tools. A controlled experiment was performed in which we addressed two types of modelling tools-desktop and cloud-based, in respect to two types of work-individual and e-collaboration. Within these treatments, we observed the productivity of 129 undergraduate IT students, who performed different types of modelling activities. The experimental participants reported no statistical significant differences between self-reported expertise with the investigated tools as well as their overall characteristics. However, they did finish individual and e-collaborative activities faster when using cloud modelling tool, where significant differences in favour of the cloud modelling tool were detected during e-collaboration. If we aggregate the results, we can argue that cloud modelling tools are comparable with desktop modelling tools during individual activities and outperform them during e-collaboration. Our findings also correlate with the related research, stating that with the use of state-of-the-art Web technologies, cloud-based applications can achieve the user experience of desktop applications.
COBISS.SI-ID: 17252886
In our research we analysed the security of the use of quantum cryptography for constrained environments like wireless sensor networks. The results of the analysis elaborate a summary about the inadequacy of use of quantum cryptography in constrained environments and presents an overview of all possible applicative challenges and problems while designing quantum-based security systems for wireless sensor networks. The results of the analysis include a list of possible applicative challenges and problems that may arise in the design of the security system intended for future WSN based on quantum cryptography. The results of research presented in detail the current problems and challenges can be a guideline for developers of security systems in the case of other specific and future IS. Other research results in the field of security protocols for constrained environments in the context of the internet of things also included several innovative security protocols that are adequately secure and efficient, and therefore tailored for such environments [COBISS.SI-ID 17757974] and [COBISS.SI-ID 18766614]. The mentioned research results are based on results in the field of authenticated key agreement protocols [COBISS.SI-ID 14779926], [COBISS.SI-ID 13646102], [COBISS.SI-ID 13379606] and [COBISS.SI-ID 13645846].
COBISS.SI-ID: 17916950
Username-password authentication scheme is becoming obsolete due to advances in password cracking technologies and bad - chosen users' passwords. We have analysed the PsychoPass method for generation of easy-to-remember passwords and proposed several improvements. These modifications enable generation of very secure passwords which are easy to remember, yet they look like totally random ones. The improved method is resilient to dictionary attack, and if used properly, renders brute force attack impractical. It represents a basis for further research of the program group members in the area of improved proactive password checker and in the area of hybrid textual-graphical passwords. In the context of security issues the PsychoPass method can be used for safe and legal outsourced analyses of medical data. A secure method for such analyses was developed in the paper »Outsourcing Medical Data Analyses: Can Technology Overcome Legal, Privacy, and Confidentiality Issues?« (COBISS.SI-ID 16928790), published in the same journal, namely in Journal of Medical Internet Research.
COBISS.SI-ID: 17522198
To synthesize the existing knowledge in the field of e-learning technology acceptance, we have conducted a systematic literature review of 42 independent papers, mostly published in major journals. Furthermore, in order to view the research context by combining and analysing the quantitative results of the reviewed research studies, a meta-analysis of the causal effect sizes between common TAM-related relationships was conducted. The main findings of this study are: (1) TAM is the most-used acceptance theory in e-learning acceptance research, and (2) the size of the causal effects between individual TAM-related factors depends on the type of user and the type of e-learning technology. The results of the meta-analysis demonstrated a moderating effect of user-related factors and technology-related factors on several evaluated causal paths. Based on knowledge gained in this study several follow-up studies were conducted in which the UTAUT acceptance model (Unified Theory of Acceptance and Use of Technology) was extended and empirically validated and evaluated [COBISS.SI-ID 22585608] in [COBISS.SI-ID 22462984].
COBISS.SI-ID: 15270166
The monograph summarizes work and results of our research group in the field of embedded systems. The primary emphasis is on systematic design of safety-critical embedded systems, which are an integral part of cyber-physical systems. Instead of the rigid design procedures, we encourage ongoing verification and ensure temporal predictability throughout the life cycle of the application development. For industry, the monograph offers a practical guideline for building safety-critical systems and gives the choices of methods and tools. On the scientific level, a new approach to the design and implementation of reliable cyber-physical systems is offered. The work on improving the dependability of the CPS has been continued in the domain of fault detection and tolerance based on re-configuration of sub-systems [COBISS.SI-ID 20179222].
COBISS.SI-ID: 11970070