Ilias Petrounias

Mining Temporal Features in Association Rules

In real world applications, the knowledge that is used for aiding decision-making is always time-varying. However, most of the existing data mining approaches rely on the assumption that discovered knowledge is valid indefinitely. People who expect to use the discovered knowledge may not know when it became valid, or whether it still is valid in the present, or if it will be valid sometime in the future. For supporting better decision making, it is desirable to be able to actually identify the temporal features with the interesting patterns or rules. The major concerns in this paper are the identification of the valid period and periodicity of patterns and more specifically association rules.

A Framework for Temporal Data Mining

Time is an important aspect of all real world phenomena. Any systems, approaches or techniques that are concerned with information need to take into account the temporal aspect of data. Data mining refers to a set of techniques for discovering previously unknown information from existing data in large databases and therefore, the information discovered will be of limited value if its temporal aspects, i.e. validity, periodicity, are not considered. This paper presents a generic definition of temporal patterns and a framework for discovering them. An architecture for the mining of such patterns is presented along with a temporal query language for extracting them from a database. As an instance of generic patterns, temporal association rules are used as examples of the proposed approach.

A neuro-fuzzy-based system for detecting abnormal patterns in wireless-capsule endoscopic images

Wireless capsule endoscopy (WCE) constitutes a recent technology in which a capsule with micro-camera attached to it, is swallowed by the patient. This paper presents an integrated methodology for detecting abnormal patterns in WCE images. Two issues are being addressed, including the extraction of texture features from the texture spectra in the chromatic and achromatic domains from each colour component histogram of WCE images and the concept of a fusion of multiple classifiers. The implementation of an advanced neuro-fuzzy learning scheme has been also adopted in this paper. The high detection accuracy of the proposed system provides thus an indication that such intelligent schemes could be used as a supplementary diagnostic tool in WCE.

A Framework for Using Web Usage Mining to Personalise E-learning

Web usage mining can contribute to finding significant educational knowledge. It can play a vital role in the personalisation aspects of any domain. We propose a framework for personalizing e-learning that necessitates careful attention towards individual learning styles. We focus on identifying learning patterns of learners and the sequence of choosing learning resources in relation to their learning styles. A prototype for an adaptive Web based course has been developed where the learning environment is modifying its behaviour to reflect learning styles.

A rule-based approach for the design and implementation of information systems

This paper presents the design, implementation and evaluation of an application following the deductive approach. The application concerns a safety critical system and for modelling the application domain two conceptual models, developed within the ESPRIT project TEMPORA, were used: the Entity Relationship Time (ERT) for the structural part and the Conceptual Rule Language (CRL) for constraints, derivation and event-action rules. The mapping from the conceptual level to the deductive DBMS platform, namely MegaLog, is described with some sample results. An overview of the system, along with a simple user interface for creating different application scenarios are presented and are followed by some statistic results of a real life case study and an evaluation of the proposed approach.

An Integrated Query and Mining System for Temporal Association Rules

In real world the knowledge used for aiding decision-making is always time varying. Most existing data mining approaches assume that discovered knowledge is valid indefinitely. Temporal features of the knowledge are not taken into account in mining models or processes. As a consequence, people who expect to use the discovered knowledge may not know when it became valid or whether it is still valid. This limits the usability of discovered knowledge. In this paper, temporal features are considered as important components of association rules for better decision-making. The concept of temporal association rules is formally defined and the problems of mining these rules are addressed. These include identification of valid time periods and identification of periodicities of an association rule, and mining of association rules with a specific temporal feature. A system has been designed and implemented for supporting the iterative process of mining temporal association rules, along with an interactive query and mining interface with an SQL-like mining language.

Language Support for Temporal Data Mining

Time is an important aspect of all real world phenomena. Any systems, approaches or techniques that are concerned with information need to take into account the temporal aspect of data. Data mining refers to a set of techniques for discovering previously unknown information from existing data in large databases and therefore, the information discovered will be of limited value if its temporal aspects, i.e. validity, periodicity, are not considered. This paper presents a generic definition of temporal patterns and a framework for discovering them. A query language for the mining of such patterns is presented in detail. As an instance of generic patterns, temporal association rules are used as examples of the proposed approach.

Power load forecasting using adaptive fuzzy inference neural networks

Load forecasting is a critical element of power system operation and planning, involving prediction of the future level of demand to serve as the basis for supply and demand planning. This includes planning for transmission and distribution facilities as well as new generation plants. This paper presents the development of a novel hybrid intelligent model and validates its prediction on the short-term electric load forecasting of the Power System of the Greek Island of Crete. The architecture and learning scheme of a novel fuzzy logic system (AFINN) implemented in the framework of a neural network is proposed. The network constructs its initial rules by clustering while the final fuzzy rule base is determined by competitive learning. The results corresponding to the minimum and maximum load time-series indicate that the proposed load forecasting model provides significantly accurate forecasts, compared to conventional neural networks models.

Representation of definite, indefinite and infinite temporal information

An information system is used for representing and managing indicative information from multiple sources describing the state of an enterprise. Most information systems model enterprises that are crisp. A crisp enterprise is one that is highly quantifiable; relationships are fixed and attributes are atomic valued. The premises for this paper are precise enterprises, where data are imperfect and multiple conflicting sources of information do exist. In such cases, information can be imperfect and/or temporal, or any possible combination of each two of them. In domains where information is perfect, all information sources are absolutely reliable. In more speculative domains, like diagnosis, information may be asserted relatively to some time intervals in which it is possibly defined and probably believed. In such domains different sources of information may be assigned different degrees of reliability. Value imperfection and/or temporal indeterminacy may cause uncertainty. Temporal information may be recurring, periodical or definite.

Efficient periodicity mining of sequential patterns in a post-mining environment

Sequential pattern mining approaches mainly deal with finding the positive behaviour of a sequential pattern that can help in predicting the next event after a sequence of events. In addition, sequential patterns may exhibit periodicity as well, i.e. during weekends 80% of people who watch a movie in cinemas will have a meal in a restaurant afterwards. This is a problem that has not been studied in the literature. To confront the problem of discovering periodicity for sequential patterns we adopt and extend a periodic pattern mining approach which has been utilised in association rule mining. However, due to the sequential/temporal nature of sequential patterns, the process of finding the periodicity of a given sequential pattern increases the complexity of the above mentioned association rule mining approach considerably. As a key attribute of any data mining strategy we provide a comprehensive and flexible problem definition framework for the above mentioned problem. Two main mining techniques are introduced to facilitate the mining process. The Interval Validation Process (IVP) is introduced to neutralise complexities which emerge due to the temporal/sequential nature of sequential patterns, whereas the Process Switching Mechanism (PSM) is devised to increase the efficiency of the mining process by only scanning relevant data-sets from the source database. The approach proposed in this paper is based on a post-mining environment, where the identification of sequential patterns from a database has already taken place.