My past academic dabblings.
Incremental Clustering of Dynamic Data Streams Using Connectivity Based Representative Points (2008)
Sebastian Lühr and Mihai Lazarescu (2009) Incremental Clustering of Dynamic Data Streams Using Connectivity Based Representative Points. In Data and Knowledge Engineering, 68(1):1-27.
Download as PDF (9.6 MiB).
Abstract: We present an incremental graph based clustering algorithm whose design was motivated by a need to extract and retain meaningful information from data streams produced by applications such as large scale surveillance, network packet inspection and financial transaction monitoring. To this end, the method we propose utilises representative points to both incrementally cluster new data and to selectively retain important cluster information within a knowledge repository. The repository can then be subsequently used to assist in the processing of new data, the archival of critical features for off-line analysis, and in the identification of recurrent patterns.
Connectivity Based Stream Clustering Using Localised Density Exemplars (2008)
Sebastian Lühr and Mihai Lazarescu (2008) Connectivity Based Stream Clustering Using Localised Density Exemplars. In Proc. Pacific-Asia Conference on Knowledge Discovery and Data Mining, volume 5012 of Lecture Notes in Artificial Intelligence, pages 983–984. Springer-Verlag.
Download as PDF (1.1 MiB; © Springer-Verlag)
Abstract: Advances in data acquisition have allowed large data collections of millions of time varying records in the form of data streams. The challenge is to effectively process the stream data with limited resources while maintaining sufficient historical information to define the changes and patterns over time. It is highly desirable to handle recurrent changes without requiring the re-learning of previously observed patterns. This paper describes an evidence-based approach that uses representative points to incrementally process stream data by using a graph based method to cluster points based on connectivity and density. Critical cluster features are archived in repositories to allow the algorithm to cope with recurrent information and to provide a rich history of relevant cluster changes if a detailed analysis of past data is required. We have applied our algorithm to both synthetic and real world data sets and present results that clearly show that our approach performs better than the current established stream mining techniques: DenStream, HPStream and CluStream.
A Visual Data Analysis Tool for Sport Player Performance Benchmarking, Comparison and Change Detection (2007)
Sebastian Lühr and Mihai Lazarescu (2007) A Visual Data Analysis Tool for Sport Player Performance Benchmarking, Comparison and Change Detection. In IEEE International Conference on Tools with Artificial Intelligence, pages 289-297. Patras, Greece.
Download as PDF (1.6 MiB)
Abstract: Sports coaches today have access to a wide variety of information sources that describe the performance of their players. However, despite this great wealth of information, most techniques used to analyse performance require a significant amount of manual processing and continue to rely heavily on input from human experts. In this paper we propose an automated approach to analyse player performance. Specifically, we propose a team benchmarking and concept drift tracking based system that (1) generates adaptive baseline player performance norms, (2) interprets player performance over different time lines and (3) identifies and describes key turning points in player performance. The concept drift technique that we describe uses a combination of overlapping data windows and decision tree based learning to process the data.
Recognition of Emergent Human Behaviour in a Smart Home: A Data Mining Approach (2007)
Sebastian Lühr, Geoff West and Svetha Venkatesh (2007) Recognition of Emergent Human Behaviour in a Smart Home: A Data Mining Approach. In Pervasive and Mobile Computing, 3(2):95-116.
Download as PDF (976 KiB)
Abstract: Motivated by a growing need for intelligent housing to accommodate aging populations, we propose a novel application of intertransaction association rule (IAR) mining to detect anomalous behaviour in smart home occupants. An efficient mining algorithm that avoids the candidate generation bottleneck limiting the application of current IAR mining algorithms on smart home data sets is detailed. An original visual interface for the exploration of new and changing behaviours distilled from discovered patterns using a new process for finding emergent rules is presented. Finally, we discuss our observations on the emergent behaviours detected in the homes of two real world subjects.
Techniques for the Discovery of Anomalous Human Behaviour in Intelligent Environments (2006)
Sebastian Lühr (2006) Techniques for the Discovery of Anomalous Human Behaviour in Intelligent Environments. PhD Thesis, Department of Computing, Curtin University of Technology.
Download as PDF (4.6 MiB)
Abstract: Motivated by a desire to create smart homes that will enable the elderly to maintain their independence for as long as possible, this thesis presents techniques for detecting abnormality in human activity observed in both laboratory and real world smart environments.
The use of stochastic models as tools for learning models of normality, with which incoming observational data from a visual tracking system can be examined, is investigated. In particular, the Hierarchical Hidden Markov Model (HHMM) is applied to the training of multi-level models of behaviour to show that the hierarchical structure of the model allows for a more expressive representation of human behaviour than is possible using flat models. The usefulness of modelling duration in models of human activity is then investigated by comparing the classification and abnormality detection performance of the Hidden Markov Model (HMM) against that of the Explicit State Duration HMM (ESD-HMM). The data sets used differ primarily in the duration of activities rather than in the ordering of the events. An extension of the ESD-HMM where the state transition times are inferred from an observation signal that has been augmented with pressure mat sensor data is then introduced. Work into this area is then concluded with results from experimentation on real world data.
A data mining technique that employs Intertransaction Association Rule (IAR) mining to discover new and changing human behaviours is then presented. The Frequent Pattern Tree (FP-Tree) and the Frequent Pattern Growth (FP-Growth) algorithm are extended for IAR mining. The resulting data structure and mining algorithm, dubbed the Extended FP-Tree (EFP-Tree) and Extended FP-Growth (EFP-Growth) respectively, are benchmarked against the First Intra Then Inter (FITI) algorithm, the existing state of the art algorithm for IAR mining. Results demonstrating that the EFP-Growth algorithm is an order of magnitude computationally more efficient than FITI are presented and discussed. The viability of emergent IAR mining as a technique for identifying unexpected behaviours in a smart home environment is affirmed with a discussion of observations made mining emergent behaviours from sensor event data recorded in the homes of two real world subjects.
Finally, a novel visual interface that enables emergent behaviours to be examined in the context of the original data is introduced. Mapping emergent IARs back into the original data space, the interface is demonstrated to allow greater insight to be gained in significantly less time than is possible by manual inspection of the sensor event log data.
Emergent Intertransaction Association Rules for Abnormality Detection in Intelligent Environments (2005)
Sebastian Lühr, Svetha Venkatesh and Geoff West (2005) Emergent Intertransaction Association Rules for Abnormality Detection in Intelligent Environments. In International Conference on Intelligent Sensors, Sensor Networks and Information Processing, pages 343-347. Melbourne, Australia.
Download as PDF (100 KiB)
Abstract: This work aims to identify anomalous behaviour of people in smart environments. We propose the use of emergent transaction mining and the use of the extended frequent pattern tree as a basis. Our experiments on two data sets demonstrate that emergent intertransaction associations are able to detect abnormality present in real world data and that both short and long term behavioural changes can be discovered. The use of intertransaction associations is shown to be advantageous in the detection of temporal association anomalies otherwise not readily detectable by traditional “market basket” intratransaction mining.
An Extended Frequent Pattern Tree for Intertransaction Association Rule Mining (2005)
Sebastian Lühr, Geoff West and Svetha Venkatesh (2005) An Extended Frequent Pattern Tree for Intertransaction Association Rule Mining. Technical Report TR-2005/1, Department of Computing, Curtin University of Technology.
Download as PDF (188 KiB, revised October 2007)
Abstract: We propose the Extended Frequent Pattern Tree (EFP-Tree) to address the problem of intertransaction association rule mining where the frequent occurrence of a large number of items results in a combinatorial explosion that limits the practical application of the existing Apriori inspired mining algorithms in a smart home environment. The EFP-Tree mining algorithm avoids candidate generation by employing a divide and conquer approach that recursively finds the set of frequent intertransaction association rules. Empirical results comparing the computational performance of the EFP-Tree with the First Intra Then Inter (FITI) algorithm on real world data from a smart home are presented. Experimental results show significant computational improvement of the EFP-Tree over FITI when a large number of rules is present in the data.
Explicit State Duration HMM for Abnormality Detection in Sequences of Human Activity (2004)
Sebastian Lühr, Svetha Venkatesh, Geoff W. West and Hung H. Bui. (2004) Explicit state duration HMM for abnormality detection in sequences of human activity. In Proc. 8th Pacific Rim Intl Conf. Artificial Intelligence, volume 3157 of Lecture Notes in Artificial Intelligence, pages 983–984. Springer-Verlag, August 2004.
Download as PDF (15 KiB, © Springer-Verlag)
Introduction: Much of the current work in human behaviour modelling concentrates on activity recognition, recognising actions and events through pose, movement, and gesture analysis. Our work focuses on learning and detecting abnormality in higher level behavioural patterns. The hidden Markov model (HMM) is one approach for learning such behaviours given a vision tracker recording observations about a person’s activity. We show how the implicit state duration in the HMM can create a situation in which highly abnormal deviation as either less than or more than the usually observed activity duration can fail to be detected and how the explicit state duration HMM (ESD-HMM) helps alleviate the problem. Duration of human activity is an important consideration if we are to accurately model a person s behavioural patterns.
Duration Abnormality Detection in Sequences of Human Activity (2004)
Sebastian Lühr, Svetha Venkatesh, Geoff West and Hung H. Bui (2004) Duration Abnormality Detection in Sequences of Human Activity. Technical Report TR-2004/02, Department of Computing, Curtin University of Technology.
Download as PDF (82 KiB)
Abstract: Activity duration is an essential element in the accurate modelling of human behaviour. The application of a standard hidden Markov Model (HMM) for the detection of abnormality in sequences of human activity can create a situation in which highly unusual duration less than or greater than the duration normally observed can fail to be detected. In this paper, we show how the application of the explicit state duration HMM can alleviate this problem, enabling us to distinguish between sequences of activity in which the order of observations is identical but the duration of activities is different and to identify the presence of abnormal activity duration. Experimental results highlight the improvement over the standard HMM for both abnormality detection and classification in certain anomalous situations.
Recognition of Human Activity Through Hierarchical Stochastic Learning (2003)
Sebastian Lühr, Hung H. Bui, Svetha Venkatesh and Geoff West (2003) Recognition of human activity through hierarchical stochastic learning. In IEEE International Conference on Pervasive Computing and Communications, pages 416–423. Texas, USA.
Download as PDF (109 KiB)
Abstract: Seeking to extend the functional capability of the elderly, we explore the use of probabilistic methods to learn and recognise human activity in order to provide monitoring support. We propose a novel approach to learning the hierarchical structure of sequences of human actions through the application of the hierarchical hidden Markov model (HHMM). Experimental results are presented for learning and recognising sequences of typical activities in a home.