Dragos Arotaritei

Web mining: a survey in the fuzzy framework

This article provides a survey of the available literature on fuzzy Web mining. The different aspects of Web mining, like clustering, association rule mining, navigation, personalization, Semantic Web, information retrieval, text and image mining are considered under the existing taxonomy. The role of fuzzy sets in handling the different types of uncertainties/impreciseness is highlighted. A hybridization of fuzzy sets with genetic algorithms (another soft computing tool) is described for information retrieval. An extensive bibliography is also included.

A Low Power Wireless Personal Area Network for Telemedicine

the current common goal in medical information technology today is the design and implementation of telemedicine solutions, which provide to patients services that enhance their quality of life. Recent technological advances in sensors, low-power integrated circuits, and wireless communications have enabled the design of low-cost, miniature, and intelligent physiological sensor modules. These modules are capable of measuring, processing, communicating one or more physiological parameters, and can be integrated into a wireless personal area network (WPAN). In this paper we present the realization of a wireless personal area network for patient monitoring, as a part of a telemedicine system, capable to measure and transmit physiological parameters such us blood pressure, heart rate, respiratory rhythm, pulse oximetry SpO2, body temperature, and three ECG leads. The use of WPAN is suitable for continuous long-time patient monitoring as a part of a diagnostic procedure, can achieve medical assistance of a chronic condition, or can be supervised during recovery from an acute event or surgical procedure. For instance, the computer-assisted rehabilitation involves unwieldy wires between sensors and monitoring device that are not very comfortable for normal activity. We use the intelligent wireless sensors, based on low power microcontrollers and RF transceivers that perform the measurements and computes the physiological parameters. Individual sensor monitors specific physiological parameter and communicates with the personal server. Personal server, in form of a wireless PDA that running a personal health monitor application, receives the information from wireless sensors, activates the alarms when the measured parameters are above the limits, and communicates periodically to the central server (part of the telemedicine system) by using WiFi or GSM/GPRS connection. The monitor reacts to potential health risks and records physiological information into a local database.

Fuzzy Voronoi diagram for disjoint fuzzy numbers of dimension two

In this paper, we introduce “fuzzy Voronoi” diagrams for fuzzy numbers of dimension two FNDT by extension of crisp Voronoi diagrams for fuzzy numbers. The fuzzy Voronoi sites are defined as fuzzy numbers of dimension two. All the operations are defined in FNDT framework. We propose an analytical approach to compute the boundaries of fuzzy Voronoi diagram for pyramidal-shape FNDT.

Mining fuzzy association rules in incomplete databases

Mining quantitative association rules is a particular subject of interest in fuzzy set application theory. However, the theory generally applies to a transactional database with no missing values. A predictive algorithm is proposed in this paper in order to extrapolate (interpolate) the unknown values. A fuzzy data mining algorithm is used to discover fuzzy association rules over the extended database with filled predictive values.

Optimization of Recurrent NN by GA with Variable Length Genotype

The gradient based learning algorithms for complex hybrid neurofuzzy architectures have a lot of local minima and a seriously time consumption complexity is involved as a consequence. Genetic Algorithms with variable length genotypes are successfully used in getting better performances for systems with complex structure or, at the same performances, a less complex structure of the system. We propose a sophisticated algorithm that solves simultaneously the optimization objectives of learning algorithms in fuzzy recurrent neural networks: both regarding the fuzzy NN performances (by its fuzzy weights matrix) and regarding the architecture (number of fully connected neurons). In this paper we developed a genetic algorithm with variable length genotypes that offers a systematic way of getting a minimal neuro-fuzzy structure satisfying the above mentioned requested performance. This advantage is not to be neglected when a complex hybrid intelligent architecture must be designed without any previous details regarding it requested architecture.

Continuous respiratory monitoring device for detection of sleep apnea episodes

Because sleep disorders are common in a significant part of the entire population with diseases of the central nervous system, continuous monitoring of respiration during sleep has an important role in early diagnosis and treatment. The paper describes the research for the design and realization of a flexible, scalable and cost-effective medical monitoring device suitable to be used with patients suffering from sleep disorders, especially obstructive sleep apnea episodes. The monitoring device contains commercially available sensors for respiratory signal measurement, a data acquisition and processing module with microcontroller, and a Tablet PC. The respiration signal is acquired by using a commercially available piezoelectric thoracic belt and processing by the microcontroller in order to compute the respiratory frequency and to detect the sleep apnea episodes. On the Tablet PC a software application displays the patients respiratory frequency and activates the alerts in the interface when an apnea episode is detected. A prototype of the described respiratory monitoring device has been developed, implemented and partially tested.

An Optimization of Data Mining Algorithms Used in Fuzzy Association Rules

The research areas on this topic have been considerably extended since the first seminal paper on data mining. Fuzzy logic plays an important role in certain aspects of data mining applications such as fuzzy clustering, discovery of fuzzy association rules, and so on. Unlike a set of crisp association rules, fuzzy association rules can create a combinatorial explosion of association rules that map linguistic data to numerical values. In this paper we propose a simple technique for limiting the creation of fuzzy association rules. In addition, the computational efficiency can be improved pruning unnecessary rules.

Recurrent algebraic fuzzy neural networks based on fuzzy numbers

A hybrid structure, recurrent algebraic fuzzy neural networks (RAFNN) using fully connected recurrent neural network architecture is proposed. The hybrid structure is based on neural network topology and fuzzy algebraic systems. All the operations are defined in the frame of fuzzy arithmetic using triangular fizzy numbers (usually non-symmetric). The experimental results demonstrate the capability of algorithm and the possibility to use successfully fuzzy numbers in recurrent architecture in order to acquire a dynamic behavior.

Adapted gradient algorithm for algebraic fuzzy neural networks

A learning algorithm based on a gradient technique is introduced for the algebraic fuzzy neural network with fuzzy weights. The fuzzy weights can be triangular fuzzy numbers (usually nonsymmetric), or trapezoidal fuzzy numbers. The network is able to map a vector of triangular (trapezoidal) fuzzy numbers into any other vector of triangular (trapezoidal) fuzzy numbers.

Hand gesture detection using a stereo camera system and simulation of movement

The implementation of a hand gesture detection using a stereo camera system is used to determine the hand position in a space described by two intersecting planes and the finger position by determining the finger tips, which leads to the classification of the hand image viewed from above into six different types. The hand movement is simulated using a 3D model implemented in Matlab programming language. In order to test the performance of the image type detection, we classify using the algorithm a dataset of 6000 images of different hand positions acquired from 20 subjects. This application can be used to control a robotic arm in real time.