Mehmed Kantardzic

The frequent wayfinding-sequence (FWS) methodology: Finding preferred routes in complex virtual environments

Advances in computing techniques, as well as the reduction in the cost of technology, have made possible the viability and spread of complex virtual environments (VEs). However, efficient navigation within these environments remains problematic for the user. Several research projects have shown that users of VEs are often disoriented and have extreme difficulty completing navigational tasks. Furthermore, there is often more than one route to get to a specified destination. Novice users often lack the spatial knowledge needed to pick an appropriate route due to the deficiency of experience with the system. A number of navigation tools such as maps, 3-D thumbnails, trails, and personal agents have been proposed. The introduction of these tools have met with some degree of success, but most researchers agree that new techniques need to be developed to aid users efficiently navigate within complex VEs. In this paper, we propose the frequent wayfinding-sequence (FWS) methodology that uses a modified sequence mining technique to discover a model of routes taken by experienced users of a VE. The model is used to build an interface that provides navigation assistance to novice users by recommending routes. We conducted both real world and simulation experiments using our methodology. Results from the real world experiment suggest that the FWS approach has the potential to improve the users navigation performance and the quality of the human-computer interaction. Our simulation studies showed that our approach is scalable, efficient, and able to find useful route models for complex VEs.

Next Generation of Data-Mining Applications

This chapter focuses on the development of an active learning approach to an image mining problem for detectingEgeria densa(a Brazilian waterweed) in digital imagery. An effective way of automatic image classification is to employ learning systems. However, due to a large number of images, it is often impractical to manually create labeled data for supervised learning. On the other hand, classification systems generally require labeled data to carry out learning. In order to strike a balance between the difficulty of obtaining labeled images and the need for labeled data, we explore an active learning approach to image mining. The goal is to minimize the task of expert labeling of images: if labeling is necessary, only those important parts of an image will be presented to experts for labeling. The critical issues are: (1) how to determine what should be presented to experts; (2) how to minimize the number of those parts for labeling; and (3) after a small number of labeled instances are available, how to effectively learn a classifier and apply it to new images. We propose to use ensemble methods for active learning in Egeria detection. Our approach is to use the combined classifications of the ensemble of classifiers to reduce the number of uncertain instances in the image classification process and thus achieve reduced expert involvement in image labeling. We demonstrate the effectiveness of our proposed system via experiments using a real-world application of Egeria detection. Practical concerns in image mining using active learning are also addressed and discussed.Trends in data-mining applications : from research labs to fortune 500 companies. 1. Mining wafer fabrication : framework and challenges. 2. Damage detection employing data-mining techniques. 3. Data projection techniques and their application in sensor array data processing. 4. An application of evolutionary and neural data-mining techniques to customer relationship management. 5. Sales opportunity miner : data mining for automatic evaluation of sales opportunity. 6. A fully distributed framework for cost-sensitive data mining. 7. Application of variable precision rough set approach to care driver assessment. 8. Discovery of patterns in earth science data using data mining. 9. An active learning approach to Egeria densa detection in digital imagery. 10. Experiences in mining data from computer simulations. 11. Statistical modeling of large-scale scientific simulation data. 12. Data mining for gene mapping. 13. Data-mining techniques for microarray data analysis. 14. The use of emerging patterns in the analysis of gene expression profiles for the diagnosis and understanding of diseases. 15. Proteomic data analysis : pattern recognition for medical diagnosis and biomarker discovery. 16. Discovering patterns and reference models in the medical domain of isokinetics. 17. Mining the cystic fibrosis data. 18. On learning strategies for topic-specific web crawling. 19. On analyzing web log data : a parallel sequence-mining algorithm. 20. Interactive methods for taxonomy editing and validation. 21. The use of data-mining techniques in operational crime fighting. 22 .Using data mining for intrusion detection. 23. Mining closed and maximal frequent itemsets. 24. Using fractals in data mining. 25 .Genetic search for logic structures in data.

User mobility oriented predictive call admission control and resource reservation for next-generation mobile networks

This paper addresses the problem of resource reservation (RR) and call admission control (CAC) in wireless mobile networks. A new user mobility oriented predictive scheme called PCAC-RR is proposed for call admission control and resource reservation. The main goal is to reduce the call dropping probability and the call blocking probability, and to maximize the bandwidth utilization. By analyzing the previous movements of the mobile users, we generate local and global mobility profiles for mobile users, which are utilized effectively in the prediction of the future path of a mobile user. Extensive simulation is used to study the performance of the proposed technique and to compare its performance with other two techniques: FR-CAT2 and PR-CAT4. Simulation results show that the proposed scheme has a significantly better performance compared to the other two schemes.

Guest Editors' Introduction: Distributed Data Mining--Framework and Implementations

It is now possible to gather and store incredible volumes of data, creating opportunities for large-scale data-driven knowledge discovery. Data mining technology has emerged as a means of performing this discovery, and the added dimension of distributed data mining increases this processs complexity.

Clustering of online game users based on their trails using self-organizing map

To keep an online game interesting to its users, it is important to know them. In this paper, in order to characterize user characteristics, we discuss clustering of online-game users based on their trails using Self Organization Map (SOM). As inputs to SOM, we introduce transition probabilities between landmarks in the targeted game map. An experiment is conducted confirming the effectiveness of the presented technique.

An extensible service oriented distributed data mining framework

This paper discusses a new approach for developing a service-oriented infrastructure for distributed data mining applications. The proposed architecture hides the complexity of implementation details and enables users to perform data mining in a utility-like fashion. The service-oriented architecture provides an autonomic data mining framework where selfdescribing data mining services can be automatically discovered on the Internet. Moreover, this structure allows for the implementation of data mining algorithms for processing data on more than one site in a distributed manner. The performance of the proposed distributed data mining framework is compared to a standard data mining approach to demonstrate its effectiveness.

The New Generation of Automatic Landmark Detection Systems: Challenges and Guidelines

Abstract Landmarks are salient objects in an environment. They play an important role in navigation by serving as orientation aids and marking decision points. Recently, there have been several efforts to design methods to automatically designate certain buildings with salient features as landmarks. All of these methodologies consist of similar steps: (a) establishing a neighborhood, usually around an intersection, (b) performing statistical or data mining analysis to find the building with outlier characteristics, and (c) establishing this salient building as the local landmark. Although these advances are significant, we believe that there are still several key issues that need to be fully addressed in order to realize the new generation of Automatic Landmark Detection Systems (ALDSs). Currently, the main shortcomings in the domain of ALDSs is the lack of a thorough and systematic study of attributes of objects that are analyzed to select landmarks, and deficient experimental verification of the benefits of ALDSs to the end users. Unless, these shortcomings are thoroughly addressed, the viability, applicability, and usefulness of ALDSs are uncertain. On the other hand, automatic landmark detection has the potential to be a dynamic, fascinating, and interdisciplinary research topic with wide applicability. Therefore, the goal of this paper is to discuss the current shortcomings in the domain of landmark detection, propose some preliminary solutions, and provide general guidelines for implementation of the new generation of ALDSs. Specifically, we discuss and promote the importance of: (a) widening the types of attributes analyzed in the landmark detection process, (b) weighting each attribute relative to its significance, (c) extending the types of objects considered as landmark candidates beyond just buildings, (d) identifying landmarks outside the vicinity of intersections, (e) identifying false landmarks along routes, and (f) using virtual environments for experiments with ALDSs. Throughout the paper, we discuss several demonstrative examples and experiments to clarify and support the ideas and concepts that are being promoted.

A grid density based framework for classifying streaming data in the presence of concept drift

Mining data streams is the process of extracting information from non-stopping, rapidly flowing data records to provide knowledge that is reliable and timely. Streaming data algorithms need to be one pass and operate under strict limitations of memory and response time. In addition, the classification of streaming data requires learning in an environment where the data characteristics might change constantly. Many of the classification algorithms presented in literature assume a 100 % labeling rate, which is impractical and expensive when data records are rapidly flowing in. In this paper, a new incremental grid density based learning framework, the GC3 framework, is proposed to perform classification of streaming data with concept drift and limited labeling. The proposed framework uses grid density clustering to detect changes in the input data space. It maintains an evolving ensemble of classifiers to learn and adapt to the model changes over time. The framework also uses a uniform grid density sampling mechanism to obtain a uniform subset of samples for better classification performance with a lower labeling rate. The entire framework is designed to be one-pass, incremental and work with limited memory to perform any-time classification on demand. Experimental comparison with state of the art concept drift handling systems demonstrate the GC3 frameworks ability to provide high classification performance, using fewer models in the ensemble and with only 4-6 % of the samples labeled. The results show that the GC3 framework is effective and attractive for use in real world data stream classification applications.

Improving Click Fraud Detection by Real Time Data Fusion

Click fraud is a type of Internet crime that occurs in pay per click online advertising when a person, automated script, or computer program imitates a legitimate user of a Web browser clicking on an ad, for the purpose of generating a charge per click without having actual interest in the target of the ads link. Most of the available commercial solutions are just click fraud reporting systems, not real-time click fraud detection and prevention systems. A new solution is proposed in this paper that will analyze the detailed user click activities based on data collected form different sources. More information about each click enables better evaluation of the quality of click traffic. We utilize the multi source data fusion to merge client side and server side activities. Proposed solution is integrated in our CCFDP V1.0 system for a real-time detection and prevention of click fraud. We have tested the system with real world data from an actual ad campaign where the results show that additional real-time information about clicks improve the quality of click fraud analysis.

Feature extraction using random matrix theory approach

Feature extraction involves simplifying the amount of resources required to describe a large set of data accurately. In this paper, we propose to broaden the feature extraction algorithms with Random Matrix Theory methodology. Testing the cross-correlation matrix of variables against the null hypothesis of random correlations, we can derive characteristic parameters of the system, such as boundaries of eigenvalue spectra of random correlations, distribution of eigenvalues and eigenvectors of random correlations, inverse participation ratio and stability of eigenvectors of non-random correlations. We demonstrate the usefullness of these parameters for network traffic application, in particular, for network congestion control and for detection of any changes in the stable traffic dynamics.