Abstract. Road intersections have become the places of high road incidents and car collisions. Our hypothesis is that a system can be made aware of dangerous situations at road intersections and warn drivers accordingly. Moreover, over time, the system can learn (or re-learn) such “patterns ” of danger for specific intersections given a history of rich collision data collected via sensors (that exist today). Based on the assumption that such a history of sensory data about colliding vehicles can be obtained, we show useful patterns that can be extracted. This paper presents our framework for intersection understanding, presenting simulated results suggesting that a fragment of the world (i.e. intersections) can be more deeply understood by mining appropriate sensor data. The simulated environment of the road intersections forming the basis of a real-world implementation and testing of the framework are discussed here. The recent results of mining traffic and collision data generated by the simulation are also included in this paper. 1

Due to the inherent characteristics of data streams, appropriate mining techniques heavily rely on window-based processing and/or (approximating) data summaries. Because resources such as memory and CPU time for maintaining such summaries are usually limited, the quality of the mining results is affected in different ways. Based on Frequent Itemset Mining and an according Change Detection as selected mining techniques, we discuss in this paper extensions of stream mining algorithms allowing to determine the output quality for changes in the available resources (mainly memory space). Furthermore, we give directions how to estimate resource consumptions based on user-specified quality requirements. 1

Developments in data streams, coupled with the growth in mobile and pervasive devices, have led to the emergence of Ubiquitous Data Mining (UDM). UDM aims to perform data stream mining in a ubiquitous environment with resource-constrained and/or mobile devices. Over the past few years, stream mining techniques have attracted the attention of the data mining community. However these techniques have not addressed the problems imposed by applying the mining technique in a ubiquitous environment. Algorithm Output Granularity (AOG) has been proposed as a generic approach to enable resource-awareness in data stream mining through adaptation. AOG has been applied to lightweight mining techniques and proved its efficiency. Due to the generality of the approach, we propose to apply AOG to an efficient stream clustering technique: Very Fast K-Means (VFKM). It is an extension of K-Means for data stream clustering. VFKM is able to deal with continuous data rather than a static dataset. In this paper, we propose and develop a resource-aware version of Very Fast K-Means to enable its operation for UDM applications. Our model for Resource-Aware Very Fast K-Means (RA-VFKM) is able to adapt to variations in memory availability on mobile devices. We have experimentally demonstrated that such an adaptation enables our RA-VFKM to converge and provide results in situations (such as critically low available memory) where VFKM tends to result in an execution failure.

Abstract. Data streams are produced continuously at a high speed. Most data stream mining techniques address this challenge by using adaptation and approximation techniques. Adapting to available resources has been addressed recently. Although these techniques ensure the continuity of the data mining process under resource limitation, the quality of the output is still an open issue. In this paper, we propose a generic model that guarantees the quality of the output while maintaining efficient resource consumption. The model works on estimating the quality of the output given the available resources. Only a subset of these resources will be used that guarantees the minimum quality loss. The model is generalized for any data stream mining technique. 1

declarations are made: I hereby declare that this thesis contains no material which has been accepted for the award of any other degree of diploma at any university or equivalent institution and that, to the best of my knowledge and belief, this thesis contains no material previously published or written by another person, except where due reference is made in the text of the thesis.

Due to the inherent characteristics of data streams, appropriate mining techniques heavily rely on window-based processing and/or (approximating) data summaries. Because resources such as memory and CPU time for maintaining such summaries are usually limited, the quality of the mining results is affected in different ways. Based on Frequent Itemset Mining and an according Change Detection as selected mining techniques, we discuss in this paper extensions of stream mining algorithms allowing to determine the output quality for changes in the available resources (mainly memory space). Furthermore, we give directions how to estimate resource consumptions based on user-specified quality requirements. 1

Abstract not found