Abstract: Matching occurrences of terms (strings) from a dictionary is one of the typical approaches for the extraction of relevant information from large collections of unstructured texts, such as electronic health records (EHRs). A common problem with dictionary-based approaches is matching terms with typographical or orthographical errors. String distance metrics can be used to alleviate this problem, by providing partial matches ranked by the similarity with the sought-out term. In this paper, we evaluate the most commonly used string distance metrics for the task of approximate dictionary matching in EHRs written in the Serbian language. Our results show that the best performing metric is the Jaro distance metric. We also report on the most frequent sources of matching errors.

Abstract: The network behavior analysis relies on the understanding of normal or acceptable behavior characteristics in the network communication, in order to efficiently detect the anomalous traffic patterns and deviations that could cause performance issues or indicate a breach, thus allowing near real-time alerting and visibility of the potential network security threats. In contrast to the signature based intrusion detection systems, this approach is extremely beneficial not only for identifying unknown threats, zero-day attacks, and suspicious behavior regardless the used cryptographic methodology, but also to identify and allow the performance optimization opportunities. We propose a comprehensive architecture for practical implementation of the flow based anomaly detection solution for real life use cases, which is based on the combination of the entropy calculation and machine learning techniques, with the ability to model the attacks and generate representative labelled training data set.

Abstract: This research relies on the proposed comprehensive flow based anomaly detection architecture, which is a complex solution that encompasses support modules for entropy calculation and for machine learning processing. The focus of this paper is on different machine learning algorithm performances in real-network scenarios. The research relies on the use of the modified CTU-13 dataset, entropy-based data preprocessing and performance analysis of a range of machine learning algorithms for modelled anomaly scenarios with synthetically generated flows. The architecture is an original solution, which is planned for further real-network application, targeting the possible support for a range of different use cases.

Abstract: — Distributed ledger technology (DLT) and blockchain, alongside with smart contracts, are seen with hope by many actors involved in international trade as a new opportunity to further facilitate and digitize international trade transactions. In order to mitigate deficiencies of the current legislation, and to avoid pitfalls that limited digitization of trade in previous decades, a carefully planned transition needs to be conducted. For this reason, we have built a solution that eases the transition to paperless trade by allowing for unobtrusive coexistence of traditional trading contracts and corresponding smart contracts running on the blockchain. Coexistence is achieved by automatically translating relevant trade terms and conditions specified in traditional trading contracts to the corresponding logic implemented in smart contracts.

Abstract: In this paper, we present a model for the automatic detection of dementia in patients based on their demographic data and the results of neurological and psychological tests. We classify patients according to four major dementia types, as defined by the domain experts. All patient features used in our model are obtained by noninvasive tests and can be easily collected by people who are not domain experts (neurologists or psychiatrists). In this way, our model could be applied to screen a much bigger population and would allow for more frequent screening of the individuals, compared to the traditional approach which requires manual diagnosis by the domain expert. In this way, we hope to alleviate the problem of early dementia diagnoses, which is critical for the patient’s quality of life. We train and evaluate our model on the publicly available Open Access Series of Imaging Studies (OASIS) dataset and obtain the f-measure of 92% and an accuracy of 94%. To the best of our knowledge, this is the best performance achieved by using only the demographic, neurological, and psychological data. It should also be noted that similar solutions classify patients in two (patients with and without dementia) or three classes (patients with mild, severe and no dementia), while we perform multi-category classification according to four major types of dementia.

Abstract: This paper describes the model that captures a variety of features important for code readability: visual, structural and textual. Code readability can be defined as a measure of how easy it is to understand the logical context of the source code, work on it collaboratively and maintain the same. Successful classification of code as readable or unreadable is a prevalent problem in today AI’s world. By the automatic discovery of unreadable code, we could significantly reduce the time needed for software development and maintenance, enforce best practices and potentially discover bugs in the code. Current solutions for the measurement of code readability are still unsatisfactory from the aspect of accuracy. To build an accurate code readability model, an appropriate dataset is needed. All existing solutions use a set of structural and/or textual features, but none of them use a visual component. Accordingly, this work represents an improvement in the described problem, introducing visual component. The visual features in our model serve to express the visual focus of the person while reading a piece of code with the hypothesis that a person will pay more attention to the more complex and penitentially less readable parts of the programming code. On the other side, the structural features are used to express the key programming concepts of the target programming language and describe the impact of code’s general structure on its readability, as well as the impact of some rarely used concepts in contrast to often used ones. Finally, the textual features, extracted from comments and identifiers, describe the semantics of software's logic and in that way contribute to a higher degree of code comprehension and readability.

Abstract: Open Big Data is steadily gaining momentum and rapidly becoming a new technology hub in industry and science that motivates technology trends to data-centric architecture and operational models. Therefore, the definition of basic information/semantic/operational models and architectural components that encompasses what is called an Open Big Data Ecosystem is becoming a necessity. The purpose of this study is to present our vision of a Big Data Ecosystem and to discuss issues observed with quality of open datasets relevant for the healthcare industry. Furthermore, the paper will propose a solution that will help the healthcare industry to take full advantage of the emerging trends in building pharmaceutical data lakes.

Abstract: In this paper, we will analyze how can we use blockchain technology for access control, that does not require trust in a third party. Current model, in which third parties collect and control massive amounts of personal data is questioned, because there are many reported incidents of security breaches that are compromising user’s personal data. On the other end, we have new EU directive, General Data Protection Regulation (GDPR), that is aiming to achieve complete protection of personal data in the EU and for the free movement and removal of such data. We describe personal data management system that is decentralized that ensures users are the owners and controllers of their data. Finally, we analyze how solution is GDPR compliant and possible future extensions.

Abstract: Testing MDE (Model Driven Development) solutions can be challenging due to their complexity and constant evolution. If the solution is used for product customization of a large scale software product line and introduced in a later phases of development, developing and maintaining testing infrastructure becomes increasingly difficult. In this paper, we examine available techniques for adequate implementation of major test types and present our approach to establishing initial test data, test cases, and validating test results for our MDE solution that supports customization of every layer of a large scale web and desktop business application and code generation of more than 150 different file types.

Abstract: The main goal of this paper is to present a novel approach of the Machine Learning (ML) in predicting the trend of the water properties within wastewater treatment processes. Bibliographical analysis shows legacy approaches the most frequently adopted, such as the Feedforward Neural Network – FNN with few layers. The ML models proposed here uses, instead, Convolutional Neural Network – CNN with several layers and a number of parameters. The use of the embeddings is also proposed to manage the categorical features and thus reach an higher performance. A real case example of application is presented, by analysing real data in a given period to prove the quality of the ML algorithm architecture designed.

Abstract: This paper describes application for remote monitoring of toxic gases in underground mines. The system accurately measures the concentration of several toxic gases: CO, CO2, CH4, H2, NH3 and other environmental parameters such as temperature and humidity. Data is transmitted through a wireless sensor network, making the system applicable where wired communication is a problem. The sensor system is portable and it can easily be moved from one place to another. The application provides adequate signaling based on the predetermined critical levels of certain gases.

Abstract: The main focus of this paper is the design of a system that will enable a declarative approach, based on a set of rules, to support complex pricing calculation logic for the insurance industry. The insurance industry is highly regulated by legal and specific domain enforced rules. Those rules are frequently adjusted, therefore creating additional effort for software maintenance, as it needs to be updated regularly as well as on short notice when sudden changes are introduced. The key feature of the proposed system is its adaptability to those frequent changes in environmental parameters in real time, achieved through the use of rule-based calculation module. This approach brings necessary flexibility to the system, reducing maintenance downtime as well as reduced susceptibility to errors during source code changes. Our approach relies on rules-based systems (engines) that execute complex business computations, declared as a set of business rules, for the insurance sector. Keywords expert systems, rule-based systems, Drools, risk management, insurance policy

Abstract: Three-dimensional (3D) reconstruction of freeform physical shapes, which are present in patient-specific dental bone grafts, can be produced by application of reverse engineering modeling. After the loss of the tooth, the effect of bone resorption or loss of bone density caused by periodontal disease or other disease or trauma can occur. Effective implantation of dental implants however requires a firm and strong bone base. This paper presents the methodology of 3D reconstruction of patient-specific dental bone grafts using reverse engineering modeling. Initially, cone-beam computer tomography (CBCT) scans of the jawbone 3D model reconstruction was used. After that, by application of additive manufacturing using binder 3D printing technologies, a physical model was made. Oral surgeon is then manually adding material similar to clay which formed the desired shape of the bone graft at the predefined area. On the next step, 3D digitization by close-range photogrammetry (CRP) was used, to get the outer geometry of the bone graft. As a result, with the use of Boolean subtraction of jawbone 3D model and CRP 3D model with designed bone graft, 3D model of completed bone graft was reconstructed.

Abstract: Technology has progressed over the last years providing solutions to much of our daily live needs. From start-ups to big companies every business is focused on developing devices and services to almost every need people have or imagine will need someday. However, when a health impairment appends to a person or a relative, it is necessary to find a customized solution that could help cope with the problem. The present research targets at developing tangible energy efficient solutions for those faced with the need for assistance to someone they care and has been diagnosed with Dementia. The conceptualization hereby presented aims at providing a personalized solution, so that family and friends can continue to carry a normal life, assisted by a device that will last battery as long as possible. For that reason, the present study aims at a simple but feasible usage of devices to be carried by a person with a yet normal life showing dementia signs that can disrupt normal routines threatens familiar’s routines and way of life.

Abstract: Technical Debt (TD) is commonly used in practice as a measure of software quality. Due to the potential overlap between software quality and software security, an interesting topic is to investigate whether TD can be used as a software security indicator as well. However, although some softwarerelated factors (e.g. software metrics) have been studied for their ability to indicate security risk in software products, no research attempts exist specifically focusing on TD. To this end, in the present study, we empirically evaluate the ability of TD to indicate security risks in software products. For this purpose, a relatively large code repository comprising 50 open-source software applications was constructed and analyzed using popular open-source static analysis tools, in order to calculate their TD and security level (i.e. vulnerability density). Subsequently, statistical analysis was employed, to assess the relationship between TD and software security. The results of the empirical study revealed a statistically significant positive and strong correlation between the TD and the vulnerability densities of the studied software products. This provides preliminary evidence for the ability of TD to be used as an indicator of software security. To the best of our knowledge, this is the first study that empirically evaluates the relationship between TD and software security.

Abstract: In this paper, the problem of routing vehicles for the collection of municipal waste is considered, which has been increasingly explored in recent years. A logistic model for municipal waste management has been presented, which was used to select the optimal routes of collecting and transporting municipal waste for a realistic example of the city of Niš in Serbia. Two metaheuristic methods - genetic algorithm (GA) and ant colony optimization (ACO) – have been used to solve the problem of waste collection vehicles routing. An analysis of the performance of applied metaheuristic algorithms and comparison of the obtained optimal solutions has been performed. As a basic measure, the total length of the route was considered when evaluating the solution. The aim of this paper is to evaluate the applicability of GA and ACO metaheuristic optimization algorithms for the problem under consideration.

Abstract: Accurate predictive models of human mandible are the most important component in maxillofacial reconstruction surgery when medical images of the whole bone are not available (due to the bone illness, fracture or an extreme traumatic bone damage, osteoporosis). Existing predictive models are based on predicting the intermediate form of the model including all of its variations based on the input data set, but the shortcomings are inaccurate prediction of shape variations and geometry of the bone outside of the input set. In this paper, the method for building predictive 3D model of the human mandible, which geometry can be changed according to the specific patient’s bone, is presented. The method is based on the Method of Anatomical Features, with the implementation of the important improvement, with the application of artificial neural networks in the prediction of human mandible geometry. This approach enables easy personalization of the model’s shape and geometry and can serve as a template for individual treatment planning.

Abstract: The paper proposes simplification and consequent cost and duration reduction of the customized production workflow, by eliminating the need for traditional structural analysis in the design of the customized product instance from the product family represented by the parametric model. It introduces the concept of so-called compiled FEA (Finite Element Analysis) model which can be used for real-time structural analysis assistance in customized product design. Compiled FEA model is actually ML (Machine Learning) model, consisting of dataset of characteristic product parameters and associated physical properties, selected ML algorithm and a set of its associated hyperparameters. The case study of creating a compiled FEA model for the case of internal orthopedic fixator is provided.

Abstract: Secure software engineering is quickly becoming the standard for software development, due to the ever-increasing number of threats and attacks to software systems. While practices such as secure coding and testing can be achieved through automated tools, security requirements engineering, and secure design are fields which heavily rely on the security expertise of software engineers. Unfortunately, this is a skill set that is both difficult to teach and learn. Recently, a framework for teaching security design analysis was developed, based on case study analysis and the hybrid flipped classroom. This paper builds on that work and presents an application of our framework, where we construct a laboratory exercise dedicated to teaching the security design analysis for repudiation threats. Through this work, we provide additional guidance for the usage of our teaching framework and outline a laboratory exercise, which can be used as part of a university course or a workshop in a corporate training program.

Abstract: Hash rate analysis can provide important insight on the processes within blockchain based systems. Since Hash functions can be solved just by usage of brute force calculation, block chain need to utilize calculating capabilities of different calculating machines ranging from CPU, GPU or ASICs. Hash rate determine level of the overall calculating power and manipulation with hash functions delivered to blockchain system. Manipulation with this values can compromise blockchain system. This work researched ASICs development and silent mining action on CryptoNight based block chain system. Empirical proofs of this action are derived and success level of mitigation action by core development team is analyzed.