The proper planning of repairs and rehabilitation of the asphalt pavement is one of the important tasks for safe driving. The most common form of distress on asphalt pavements are potholes, which can compromise safety, and result in vehicle damage. Timely repairing potholes is crucial in ensuring the safety, quality of driving, and reducing the cost of vehicle maintenance. Many of the existing methods for pothole detection often use sophisticated equipment and algorithms, which require substantial amount of data for filtering and training. Consequently, as a result of intensive computational processing, this can lead to long execution time and increased power consumption. In this paper, we propose an efficient unsupervised vision-based method for pothole detection without the process of training and filtering. Our method first extracts asphalt pavements by analysing RGB color space and performing image segmentation. When the asphalt pavement is detected, the search continues in detected region only. The method is tested on online image data set captured from different cameras and angles, with different irregular shapes and number of potholes. The results indicate that the method is suitable as a pre-processing step for other supervised methods.

This paper presents results of an analysis of the impact of the human resources changes in Scrum teams. Four Scrum teams were tracked (two developments and two quality assurance) along with their productivity and performance. Analysis showed that human resources changes have a significant impact on the entire team and its behavior. Their effort increased by adding overtime hours. In the same time, their performance and effective work decreased, which is reflected on the quantity of work that can be billed to the client. The analysis shows that it takes, in average, three sprints (each lasting fourteen days) for new team members to fully adjust to the team development process and acquire a business knowledge needed for maximum productivity. Teams whose members have been working together longer period and who have more senior members can adjust to team shifts more quickly. The analysis also showed a correlation between quality assurance and development team - when development team had extra utilization due to overtime, quality assurance team had an increase in overtime hours almost proportionately.

This paper is related to a research of modelling fuzzy-neural systems using the coevolutionary algorithm, and has the focus on advantages of using the coevolutionary algorithm for system structure optimization. In the context of this work, the term fuzzy-neural system defines the system that can be used as the fuzzy system with all its functionalities or as the neural network with all its functionalities. The hybridization of fuzzy logic, neural networks and coevolutionary algorithm and its architecture are presented in general, and the role of the coevolutionary algorithm in structure optimization is described in details. Results of testing with Iris Database, from UCI Machine Learning Repository are also presented. Tests performed during the research supports the conclusion that usage of the coevolutionary algorithm for the fuzzy-neural system’s structure optimization is very efficient.

The water level in an artificial lake is important not only for the production of electric energy but also for other activities such as tourism, irrigation and drought control. The water level in the lake is influenced by various factors, among which the most important include: the water inflow, discharge of water and water seepage. In this research, artificial neural networks (ANN) are selected for the water level prediction because of their well-known abilities for learning from examples. A total of 29 years of water level measurement data was used for ANN training and validation. This paper represents a sequential approach for the short-term water level prediction in Jablanicko lake by using only water level data. With regard to sequential approach for every step of the prediction, the most recent data were used for ANN training. Two types of ANNs were used in this study: Nonlinear Autoregressive (NAR) neural networks and Feed Forward Back Propagation (FFBP) neural networks. The main focus of this study was on NAR networks prediction of water level, while FFBP networks were used for comparison purposes. The results showed that neural networks can provide quality water level prediction even if only water level data is used.

Customization of the existing software for a new customer can become problematic if negative factors are not controlled and prevented. This paper discusses customization of software product line (SPL) for an unprepared environment and emphasis importance of identifying this risk factor. The SPL can be viewed as a set of applications for similar environments that share many common elements (processes, interfaces, data models, etc.), but it is important that each instance in the SPL application family is also viewed as a consistent and stabile system that has some specific properties, comparing to others. SPL customization - implementing those specific properties should not endanger stability of the common elements and the system as a whole. Very often preparation for the SPL customization is not as serious as for developing a new software product. That can cause SPL customization for a pretty unknown environment which can be unprepared for accepting such software. Environment that is not willing or not capable to accept (or understand) software in its full potential is named unprepared environment, in this paper. Problems that can arise in an unprepared environment can cause that implemented system is unstable or improperly used, which is described through the case study.

In today's global business environment, the importance of customer service, cost-competitiveness, and quality are key factors in determining an organization's success, or undesirable failure. Organizations try to optimize their processes to maximize their profits and make the very process faster. Users usually work with documents in the process. Working with documents makes the process more slowly, since the documents are important to be scanned and attached to the form. This work presents the optimization of such processes. It is achieved by automatic integration of Business Process Management and Electronic Document Management Systems. Improvements and results achieved by proposed integration are presented in this research. The model that is created enables monitoring of defined Key Performance Indicators in the identification process of the bottlenecks in the process. The process can be optimized by increasing the number of resources on the activities that are a bottleneck in the process. Such a solution has been tested in the process of opening a bank account.

Modern computer network IDS (Intrusion Detection Systems) and IPS (Intrusion Prevention Systems) increasingly use SOM (Self-Organizing Map) in classification of network traffic process. In this research, training data preparation is implemented as a preprocessing block composed of SOMs. Several networks with different characteristics are linked cascade and parallel for the purpose of creating SOM block. This block is used for reduction of training data through process of clustering data in smaller subsets. Training data are divided in clusters and used for training of ANFIS (Adaptive Network Based Inference System) components of the system. IDS hybrid structure consists of SOM block cascade linked with fuzzy system. The proposed hybrid structure is trained, tested and validated using KDD CUP 99 data set. This paper presents advantages and disadvantages of hybrid approach based on neural networks and fuzzy logic comparing it to similar solutions that can be found in the literature. New refined training data set is prepared with proposed solution. Overall classification using this solution, gives better results for one class compared to best results of KDD CUP 99 competition and other recently developed solutions.

The choice of the first programming language and the corresponding programming paradigm is an important part of the software development process. Knowing the advantages and constraints of individual programming paradigms is important as it can be crucial for successful software implementation. In this paper we conduct an empirical comparison of functional and object-oriented programming languages using analog examples in C#, F#, Haskell, and Java. Three algorithms were implemented: algorithm for solving N queens problem, algorithm for generating n-th left-truncatable prime and merge sort algorithm in C#, F#, Haskell and Java programming languages. An overview of programming languages efficiency is given by measuring two basic parameters: number of lines of code and program execution speed. Also, system resource usage is monitored during execution. Limited experiments showed that the programming language Java is faster than the other three languages whose performances were measured. Java was surprisingly fast on these problems that are more suitable for functional programming languages. Haskell was less memory intensive (up to two times less than Java) with similar execution times, while .NET languages were slower up to four times in comparison to Java. Object-oriented languages C# and Java had significantly more lines of code for all three algorithms when compared to functional programming language Haskell and the hybrid one F#.

Presents information on the 25th International Conference on Information, Communication, and Automation Technologies.

Component-based software engineering ensures high software quality and rapid application development. However many technological changes are observed since early outbreak of component-based software engineering (CBSE) paradigm on 1990's. One of these changes is clear shift towards mobile and cloud computing platforms. The goal of this paper is to outline major changes of the component concept and changes of CBSE in general influenced by the mobile platforms. The models of iOS, Android and Windows Phone Applications are described. The paper discloses main similarities and differences of the legacy systems and the mobile component-based systems.

There is an increasing number of diabetic diseases in the population. Early diagnosis increases possibility of healing and decreases healing expenses. This paper is focused on modeling of an expert system for assessing diabetes risk using artificial intelligence methods. The model is hierarchical with neuro-fuzzy blocks and voting on the output. Model optimization (learning) is done using the data of real patients acquired in a one public health institution. The expert system is implemented in the Matlab/Simulink environment. Validation of the system shows high sensitivity of 100% which is important for early diagnoses. Specificity is lower - only 90% which means that some patients are sent to further diagnoses although they are healthy. Results indicate that this modeling approach is applicable on assessing diabetes risk. The medical domain knowledge and experience contained in the real data is successfully transferred in the solution model - the proposed expert system. Validation of the proposed expert system indicates that such a system can be used as an auxiliary expert for early diagnoses of diabetes and improve the quality of health systems with lower diagnostic expenses.

This paper presents novel GA-ANFIS expert system prototype for dermatological disease detection by using dermatological features and diagnoses collected in real conditions. Nine dermatological features are used as inputs to classifiers that are based on Adaptive Neuro-Fuzzy Inference Systems (ANFIS) for the first level of fuzzy model optimization. After that, they are used as inputs in Genetic Algorithm (GA) for the second level of fuzzy model optimization within GA-ANFIS system. GA-ANFIS system performs optimization in two steps. Modelling and validation of the novel GA-ANFIS system approach is performed in MATLAB environment by using validation set of data. Some conclusions concerning the impacts of features on the detection of dermatological diseases were obtained through analysis of the GA-ANFIS. We compared GA-ANFIS and ANFIS results. The results confirmed that the proposed GA-ANFIS model achieved accuracy rates which are higher than the ones we got by ANFIS model.

Many different applications like face/people detection, image content interpretation, de-identification for privacy protection in multimedia content, etc. requires skin detection as a pre-processing step. There is no a perfect solution for skin detection, since this process is a compromise on speed, simplicity and precision (detection quality). There are many different techniques for skin detection modeling ranging from simple models based on one or several thresholds to advanced models based on neural network, Bayesian classifier, maximum entropy, k-means clustering, etc. This paper proposes a simple model, based on ratios of red, green and blue components of the RGB color model. It describes how to make a compromise in a skin detection modeling by using three levels of rules. Data analysis that supports conclusions is performed on the dataset from Universidad de Chile (UChile, dbskin2 - complete set) that contains 103 images and their annotations.

Clustering is a crucial step in the analysis of gene expression data. Its goal is to identify the natural clusters and provide a reliable estimate of the number of distinct clusters in a given data set. In this paper we propose new hybrid algorithm for clustering of microarray data based on spectral clustering and k-means. Our algorithm consist of four steps, including preprocessing or filtering step, and finding optimal number of clusters by using two different clustering methods based on hierarchical and partition-based approaches. Then, we cluster data based on similarity/dissimilarity metrics with spectral clustering. In the final step, we select centroid genes based on kmeans results. The proposed method was tested on six data sets from GEMS microarray database. When compared with existing single or combination of clustering methods, our results indicate about 10% improvement in selection of representative genes.