Marwan Hassani

Density-Based projected clustering of data streams

In this paper, we have proposed, developed and experimentally validated our novel subspace data stream clustering, termed PreDeConStream. The technique is based on the two phase mode of mining streaming data, in which the first phase represents the process of the online maintenance of a data structure, that is then passed to an offline phase of generating the final clustering model. The technique works on incrementally updating the output of the online phase stored in a micro-cluster structure, taking into consideration those micro-clusters that are fading out over time, speeding up the process of assigning new data points to existing clusters. A density based projected clustering model in developing PreDeConStream was used. With many important applications that can benefit from such technique, we have proved experimentally the superiority of the proposed methods over state-of-the-art techniques.

Towards a Mobile Health Context Prediction: Sequential Pattern Mining in Multiple Streams

Context prediction is an emerging topic in the fields of data mining and information management which is both promising and challenging. Predicting the location of mobile objects was a frequently tackled subtask of mobile context prediction in recent researches. For scenarios of managing health information of mobile persons, the prediction of near future health status of persons is at least equally important to predicting their location. We introduce in this paper, to the best of our knowledge, a first method for predicting a next health context of mobile persons equipped with body sensors and a mobile device. The suggested Prefix Span-based method searches for sequential patterns within multiple streaming inputs from the body sensors as well as other contextual streams that influence the health context. We discuss additionally the implementation of our method in an energy aware mobile-server environment.

EDISKCO: energy efficient distributed in-sensor-network k-center clustering with outliers

Clustering is an established data mining technique for grouping objects based on similarity. For sensor networks one aims at grouping sensor measurements in groups of similar measurements. As sensor networks have limited resources in terms of available memory and energy, a major task sensor clustering is efficient computation on sensor nodes. As a dominating energy consuming task, communication has to be reduced for a better energy efficiency. Considering memory, one has to reduce the amount of stored information on each sensor node. For in-network clustering, k-center based approaches provide k representatives out of the collected sensor measurements. We propose EDISKCO, an outlier aware incremental method for efficient detection of k-center clusters. Our novel approach is energy aware and reduces amount of required transmissions while producing high quality clustering results. In thorough experiments on synthetic and real world data sets, we show that our approach outperforms a competing technique in both clustering quality and energy efficiency. Thus, we achieve overall significantly better life times of our sensor networks.

Adaptive Multiple-Resolution Stream Clustering

Stream data applications have become more and more prominent recently and the requirements for stream clustering algorithms have increased drastically. Due to continuously evolving nature of the stream, it is crucial that the algorithm autonomously detects clusters of arbitrary shape, with different densities, and varying number of clusters. Although available density-based stream clustering are able to detect clusters with arbitrary shapes and varying numbers, they fail to adapt their thresholds to detect clusters with different densities. In this paper we propose a stream clustering algorithm called HASTREAM, which is based on a hierarchical density-based clustering model that automatically detects clusters of different densities. The density thresholds are independently adapted to the existing data without the need of any user intervention. To reduce the high computational cost of the presented approach, techniques from the graph theory domain are utilized to devise an incremental update of the underlying model. To show the effectiveness of HASTREAM and hierarchical density-based approaches in general, several synthetic and real world data sets are evaluated using various quality measures. The results showed that the hierarchical property of the model was able to improve the quality of density-based stream clusterings and enabled HASTREAM to detect streaming clusters of different densities.

Precise anytime clustering of noisy sensor data with logarithmic complexity

Clustering of streaming sensor data aims at providing online summaries of the observed stream. This task is mostly done under limited processing and storage resources. This makes the sensed stream speed (data per time) a sensitive restriction when designing stream clustering algorithms. Additionally, the varying speed of the stream is a natural characteristic of sensor data, e.g. changing the sampling rate upon detecting an event or for a certain time. In such cases, most clustering algorithms have to heavily restrict their model size such that they can handle the minimal time allowance. Recently the first anytime stream clustering algorithm has been proposed that flexibly uses all available time and dynamically adapts its model size. However, the method was not designed to precisely cluster sensor data which are usually noisy and extremely evolving. In this paper we detail the LiarTree algorithm that provides precise stream summaries and effectively handles noise, drift and novelty. We prove that the runtime of the LiarTree is logarithmic in the size of the maintained model opposed to a linear time complexity often observed in previous approaches. We demonstrate in an extensive experimental evaluation using synthetic and real sensor datasets that the LiarTree outperforms competing approaches in terms of the quality of the resulting summaries and exposes only a logarithmic time complexity.

Efficient Process Discovery From Event Streams Using Sequential Pattern Mining

Process mining is an emerging research area that applies the well-established data mining solutions to the challenging business process modeling problems. Mining streams of business processes in the real time as they are generated is a necessity to obtain an instant knowledge from big process data. In this paper, we introduce an efficient approach for exploring and counting process fragments from a stream of events to infer a process model using the Heuristics Miner algorithm. Our novel approach, called StrProM, builds prefix-trees to extract sequential patterns of events from the stream. StrProM uses a batch-based approach to continuously update and prune these prefix-trees. The final models are generated from those trees after applying a novel decaying mechanism over their statistics. The extensive experimental evaluation demonstrates the superiority of our approach over a state-of-the-art technique in terms of execution time using a real dataset, while delivering models of a comparable quality.

Subspace clustering of data streams: new algorithms and effective evaluation measures

Nowadays, most streaming data sources are becoming high dimensional. Accordingly, subspace stream clustering, which aims at finding evolving clusters within subgroups of dimensions, has gained a significant importance. However, in spite of the rich literature of subspace and projected clustering algorithms on static data, only three stream projected algorithms are available. Additionally, existing subspace clustering evaluation measures are mainly designed for static data, and cannot reflect the quality of the evolving nature of data streams. On the other hand, available stream clustering evaluation measures care only about the errors of the full-space clustering but not the quality of subspace clustering. In this article we present a method for designing new stream subspace and projected algorithms. We propose also, to the first of our knowledge, the first subspace clustering measure that is designed for streaming data, called SubCMM: Subspace Cluster Mapping Measure. SubCMM is an effective evaluation measure for stream subspace clustering that is able to handle errors caused by emerging, moving, or splitting subspace clusters. Additionally, we propose a novel method for using available offline subspace clustering measures for data streams over the suggested new algorithms within the Subspace MOA framework.

Spatiotemporal Similarity Search in 3D Motion Capture Gesture Streams

The question of how to model spatiotemporal similarity between gestures arising in 3D motion capture data streams is of major significance in currently ongoing research in the domain of human communication. While qualitative perceptual analyses of co-speech gestures, which are manual gestures emerging spontaneously and unconsciously during face-to-face conversation, are feasible in a small-to-moderate scale, these analyses are inapplicable to larger scenarios due to the lack of efficient query processing techniques for spatiotemporal similarity search. In order to support qualitative analyses of co-speech gestures, we propose and investigate a simple yet effective distance-based similarity model that leverages the spatial and temporal characteristics of co-speech gestures and enables similarity search in 3D motion capture data streams in a query-by-example manner. Experiments on real conversational 3D motion capture data evidence the appropriateness of the proposal in terms of accuracy and efficiency.

Subspace MOA: Subspace Stream Clustering Evaluation Using the MOA Framework

Most available static data are becoming more and more high-dimensional. Therefore, subspace clustering, which aims at finding clusters not only within the full dimension but also within subgroups of dimensions, has gained a significant importance. Recently, OpenSubspace framework was proposed to evaluate and explorate subspace clustering algorithms in WEKA with a rich body of most state of the art subspace clustering algorithms and measures. Parallel to it, MOA (Massive Online Analysis) framework was developed also above WEKA to provide algorithms and evaluation methods for mining tasks on evolving data streams over the full space only.

Differential private trajectory protection of moving objects

Location privacy and security of spatio-temporal data has come under high scrutiny in the past years. This has rekindled enormous research interest. So far, most of the research studies that attempt to address location privacy are based on the k-Anonymity privacy paradigm. In this paper, we propose a novel technique to ensure location privacy in stream and non-stream mobility data using differential privacy. We portray incoming stream or non-stream mobility data emanating from GPS-enabled devices as a differential privacy problem and rigorously define a spatio-temporal sensitivity function for a trajectory metric space. Privacy is achieved through path perturbation in both the space and time domain. In addition, we introduce a new notion of Nearest Neighbor Anchor Resource to add more contextual meaning in the face of uncertainty to the perturbed trajectory path. Unlike k-Anonymity techniques that require more mobile objects to achieve strong anonymity; we show that our approach provides stronger privacy even for a single moving mobile object, outliers or mobile objects in sparsely populated regions.