Murlikrishna Viswanathan

Development of incident detection model using neuro-fuzzy algorithm

This research aims at model development for incident detection and travel time estimation using a neuro-fuzzy algorithm. Traffic incidents such as accidents, weather and construction, are a major cause of congestion. Thus incident detection and optimal travel time estimation is required for improving general traffic conditions. Until recently, two approaches related to the above were the aim of many studies. One idea is to estimate travel time using data fusion from many sources while another is to estimate optical path through travel time data. As a first step, in this paper we develop an initial model for incident detection using a neuro-fuzzy algorithm. In our experiments we find that our proposed model has a incident detection rate (DR) of over 83% and a false alarm rate (FAR) under 24%. The test results also suggest that the proposed model enhances accuracy of incident detection in an arterial road and we expect the proposed model to contribute to formal traffic policy.

A Distributed Data Mining System for a Novel Ubiquitous Healthcare Framework

Recent years have shown impressive growth in the development of ubiquitous healthcare (u-Healthcare) systems which aim for the next generation in e-Health services and associated research. Such systems are primarily being designed to provide emergency and preventive healthcare to citizens anywhere/anytime using wired and wireless mobile technologies. Data mining is an indispensable aspect of such systems and represents the process of analyzing databases to extract hidden knowledge and relationships. This paper introduces and studies the development framework of a prototype ubiquitous healthcare system initiated by the South Korean government. Preliminary results with a distributed data mining system are presented in the context of a larger future integration with the ubiquitous healthcare framework.

A hybrid soft computing approach to link travel speed estimation

The field of dynamic vehicle routing and scheduling is growing at a strong pace nowadays due to the many potential applications in urban traffic management. In recent times there have been many attempts to estimate the vehicle travel times over congested links. As opposed to the previous decade where traffic information was collected mainly by fixed devices with high maintenance costs, the advent of GPS has resulted in data being progressively collected using probe cars equipped with GPS-based communication modules. Typically traditional methods used for analyzing the data collected using fixed devices need to be extended. The aim of this research is to propose a hybrid method for estimating the optimal link speed using the data acquired from probe cars using combination of the fuzzy c-means (FCM) algorithm with multiple regression analysis. The paper describes how the probe data are analyzed and automatically classified into three groups of speed patterns and the link speed is predicted from these clusters using multiple regression. In performance tests, the proposed method is robust and is able to provide accurate travel time estimates.

Emotional-speech recognition using the neuro-fuzzy network

Emotion recognition based on a speech signal is one of the intensively studied research topics in the domains of human-computer interaction and affective computing. The presented paper is concerned with emotional-speech recognition based on the neuro-fuzzy network with a weighted fuzzy membership function (NEWFM). NEWFM has a feature selection method and makes fuzzy classifiers. In this paper, NEWFM was utilized for classifying four kinds of emotional-speech signals. This NEWFM classification method achieves as high as 86% overall classification accuracy. Significantly, the NEWFM classifier efficiently detects sadness, with a 97.5% recognition rate.

Performance analysis of dynamic mobility management for proxy mobile IPv6 networks

Proxy Mobile IPv6 (PMIPv6) is a network-based mobility management protocol that does not require mobile node involvement in mobility management. In PMIPv6, the mobile access gateway (MAG) incurs a high signaling cost to update the location of a mobile node to the remote local mobility anchor (LMA) if it moves frequently. It may cause excessive signaling traffic and increase a high traffic load on the LMA. Therefore, we propose a new mobility management scheme in proxy mobile IPv6 networks with dynamic paging support. To minimize the signaling overhead, in the proposed scheme, an idle mobile node does not register when moving within a paging area. Moreover, the size of the paging area is determined dynamically according to the changes of in the mobility and traffic patterns of the mobile node. An analytic model is applied to determine the optimal size of the paging area. The cost analysis using random walk on a connected graph model presented in this paper shows that the proposed scheme can achieve superior performance much better than of the PMIPv6 scheme.

Distributed data mining on clusters with bayesian mixture modeling

Distributed Data Mining (DDM) generally deals with the mining of data within a distributed framework such as local area and wide area networks. One strong case for DDM systems is the need to mine for patterns in very large databases. This requires mandatory partitioning or splitting of databases into smaller sets which can be mined locally over distributed hosts. Data Distribution implies communication costs associated with the need to combine the results from processing local databases. This paper considers the development of a DDM system on a cluster. In specific we approach the problem of data partitioning for data mining. We present a prototype system for DDM using a data partitioning mechanism based on Bayesian mixture modeling. Results from comparison with standard techniques show plausible support for our system and its applicability.

Knowledge-based compliance management systems - methodology and implementation

In recent times, a challenging problem for organisations worldwide has been the management of the growing number of rules, procedures, policies, and reporting requirements governing their businesses, operations, and industry. This paper considers the task of building automated knowledge-based compliance management systems. In this paper, we aim to highlight the common weaknesses found in compliance methodologies in practice. The requirements of a good research methodology for compliance are discussed. Finally, the paper presents the development of a research methodology for compliance and its validation through a progressive case study in International Transfer Pricing.

Neuro-fuzzy learning for automated incident detection

Traffic incidents such as vehicle accidents, weather and construction works are a major cause of congestion. Incident detection is thus an important function in freeway and arterial traffic management systems. Most of the large scale and operational incident detection systems make use of data collected from inductive loop detectors. Several new approaches, such as probe vehicles and video image processing tools, have recently been proposed and demonstrated. This research aims at model development for automatic incident detection and travel time estimation employing neuro-fuzzy techniques. As a first step, in this paper we develop an initial model for incident detection using a standard neuro-fuzzy algorithm. In subsequent development we propose a model where the fuzzy rules are themselves extracted from the data using an associative data mining algorithm. The results of the initial experiments suggest that the proposed model has plausible incident detection rates and low false alarm rates. The test results also suggest that the proposed model enhances accuracy of incident detection in an arterial road and can be expected to contribute to formal traffic policy.

Sketch map generation method using leveled spatial indexing technique in a mobile environment

This paper deals with the efficiency of route services in mobile environments with necessitate methods which reduce map description data and increase users understanding of the map. In order to reduce the size of map data, first, the data is aligned by importance and incrementally serviced to the user by importance. In addition, the size of the map data can be reduced by using symbols which represent data in a concise manner. Second, to increase the users understanding about route (optimal path showing start and destination) and its surroundings, we propose a new method which converts a complex route and its surroundings into a simplified sketch map. A process is also suggested to correct distortion which occurs during the sketch map generation process. For speedy service in mobile environments, a new indexing method, leveled spatial indexing, is proposed in this paper. Current spatial indexing methods do not support all the map generalization operations and real-time processing. The indexing method proposed in this paper supports all the map generalization operations and manipulates zoom in-out quickly by leveling the generalized data. To verify the efficiency of the methods proposed in this paper, several experiments using cellular phones are conducted and the results of performance evaluation are presented.

Pose-invariant face recognition using deformation analysis

Over the last decade or so, face recognition has become a popular area of research in computer vision and one of the most successful applications of image analysis and understanding. In addition, recognition of faces under varied poses has been a challenging area of research due to the complexity of pose dispersion in feature space. This paper presents a novel and robust pose-invariant face recognition method. In this approach, first, the facial region is detected using the TSL color model. The direction of face or pose is estimated using facial features and the estimated pose vector is decomposed into X-Y-Z axes. Second, the input face is mapped by a deformable template using these vectors and the 3D CANDIDE face model. Finally, the mapped face is transformed to the frontal face which appropriates for face recognition by the estimated pose vector. Through the experiments, we come to validate the application of face detection model and the method for estimating facial poses. Moreover, the tests show that recognition rate is greatly boosted through the normalization of the poses.