Kenneth McGarry

A survey of interestingness measures for knowledge discovery

It is a well-known fact that the data mining process can generate many hundreds and often thousands of patterns from data. The task for the data miner then becomes one of determining the most useful patterns from those that are trivial or are already well known to the organization. It is therefore necessary to filter out those patterns through the use of some measure of the patterns actual worth. This article presents a review of the available literature on the various measures devised for evaluating and ranking the discovered patterns produced by the data mining process. These so-called interestingness measures are generally divided into two categories: objective measures based on the statistical strengths or properties of the discovered patterns and subjective measures that are derived from the users beliefs or expectations of their particular problem domain. We evaluate the strengths and weaknesses of the various interestingness measures with respect to the level of user integration within the discovery process.

Data mining using rule extraction from Kohonen self-organising maps

The Kohonen self-organising feature map (SOM) has several important properties that can be used within the data mining/knowledge discovery and exploratory data analysis process. A key characteristic of the SOM is its topology preserving ability to map a multi-dimensional input into a two-dimensional form. This feature is used for classification and clustering of data. However, a great deal of effort is still required to interpret the cluster boundaries. In this paper we present a technique which can be used to extract propositional IF..THEN type rules from the SOM network’s internal parameters. Such extracted rules can provide a human understandable description of the discovered clusters.

Knowledge extraction from radial basis function networks and multilayer perceptrons

This paper deals with an evaluation and comparison of the accuracy and complexity of symbolic rules extracted from radial basis function networks and multilayer perceptrons. Here we examine the ability of rule extraction algorithms to extract meaningful rules that describe the overall performance of a particular network. In addition, the paper also highlights the suitability of a specific neural network architecture for particular classification problems. The study carried out on the extracted rule quality and complexity also has a direct bearing on the use of rule extraction algorithms for data mining and knowledge discovery.

Image classification using hybrid neural networks

Use of semantic content is one of the major issues which needs to be addressed for improving image retrieval effectiveness. We present a new approach to classify images based on the combination of image processing techniques and hybrid neural networks. Multiple keywords are assigned to an image to represent its main contents, i.e. semantic content. Images are divided into a number of regions and colour and texture features are extracted. The first classifier, a self-organising map (SOM) clusters similar images based on the extracted features. Then, regions of the representative images of these clusters were labeled and used to train the second classifier, composed of several support vector machines (SVMs). Initial experiments on the accuracy of keyword assignment for a small vocabulary are reported.

Qualitative evaluation of automatic assignment of keywords to images

In image retrieval, most systems lack user-centred evaluation since they are assessed by some chosen ground truth dataset. The results reported through precision and recall assessed against the ground truth are thought of as being an acceptable surrogate for the judgment of real users. Much current research focuses on automatically assigning keywords to images for enhancing retrieval effectiveness. However, evaluation methods are usually based on system-level assessment, e.g. classification accuracy based on some chosen ground truth dataset. In this paper, we present a qualitative evaluation methodology for automatic image indexing systems. The automatic indexing task is formulated as one of image annotation, or automatic metadata generation for images. The evaluation is composed of two individual methods. First, the automatic indexing annotation results are assessed by human subjects. Second, the subjects are asked to annotate some chosen images as the test set whose annotations are used as ground truth. Then, the system is tested by the test set whose annotation results are judged against the ground truth. Only one of these methods is reported for most systems on which user-centred evaluation are conducted. We believe that both methods need to be considered for full evaluation. We also provide an example evaluation of our system based on this methodology. According to this study, our proposed evaluation methodology is able to provide deeper understanding of the systems performance.

CLAIRE: A modular support vector image indexing and classification system

Many users of image retrieval systems would prefer to express initial queries using keywords. However, manual keyword indexing is very time-consuming. Therefore, a content-based image retrieval system which can automatically assign keywords to images would be very attractive. Unfortunately, it has proved very challenging to build such systems, except where either the image domain is restricted or the keywords relate only to low-level concepts such as color. This article presents a novel image indexing and classification system, called CLAIRE (CLAssifying Images for REtrieval), composed of one image processing module and three modules of support vector machines for color, texture, and high-level concept classification for keyword assignment. The experimental prototype system described here assigns up to five keywords selected from a controlled vocabulary of 60 terms to each image. The system is trained offline by 1639 examples from the Corel stock photo library. For evaluation, five judges reviewed a sample of 800 unknown images to identify which automatically assigned keywords were actually relevant to the image. The system proved to have an 80% probability to assign at least one relevant keyword to an image.

Automated trend analysis of proteomics data using an intelligent data mining architecture

Proteomics is a field dedicated to the analysis and identification of proteins within an organism. Within proteomics, two-dimensional electrophoresis (2-DE) is currently unrivalled as a technique to separate and analyse proteins from tissue samples. The analysis of post-experimental data produced from this technique has been identified as an important step within this overall process. Some of the long-term aims of this analysis are to identify targets for drug discovery and proteins associated with specific organism states. The large quantities of high-dimensional data produced from such experimentation requires expertise to analyse, which results in a processing bottleneck, limiting the potential of this approach. We present an intelligent data mining architecture that incorporates both data-driven and goal-driven strategies and is able to accommodate the spatial and temporal elements of the dataset under analysis. The architecture is able to automatically classify interesting proteins with a low number of false positives and false negatives. Using a data mining technique to detect variance within the data before classification offers performance advantages over other statistical variance techniques in the order of between 16 and 46%.

Quantifying multifaceted needs captured at the point of care. Development of a Disabilities Terminology Set and Disabilities Complexity Scale.

To develop a Disabilities Terminology Set and quantify the multifaceted needs of disabled children and their families in a district disability clinic population.

Recent trends in knowledge and data integration for the life sciences

The bioscience field has seen some spectacular advances in genomic and proteomic technologies that are able to deliver vast quantities of information on cellular activity. Such technologies are of critical importance to biology, medical science and in drug discovery. However, living systems are highly complex and to fully exploit these technologies requires knowledge at many different levels. Information such as genome sequence data, gene expression data, protein-to-protein interactions and metabolic pathways is required to understand the complexity of biological processes. The challenge for bioinformatics is to tackle the problem of fragmentation of knowledge by integrating the many sources of heterogeneous information into a coherent entity. Another problem is that the high level of biological complexity and the fragmented nature of biological research has meant that it is difficult to keep fully conversant with the latest research and discoveries. Progress in one area of biology may have implications for other areas but the dissemination of this knowledge is not straightforward; difficulties such as differences in naming conventions for genes and biological processes has led to confusion and the lack of productivity. This paper reviews the most recent research to overcome the fragmentation problem where technologies such as text mining and ontologies are used within the knowledge discovery process and the specific technical challenges they address.

Analysis of Rules Discovered by the Data Mining Process

This paper describes how symbolic rules may be extracted from Radial Basis Function neural networks and shows how they can be used by the data mining and knowledge discovery process. Rule extraction overcomes a major disadvantage of neural networks which is concerned with making the comprehensibilty of the learned internal model more open to scrutiny. Having extracted the symbolic rules we show how they are assessed and ranked for interesting or novel features. Two such techniques are presented here, the first is a data driven approach that uses objective mathematical measures to identify interesting patterns or features. The second is a goal driven method that uses subjective measures obtained from the user. The measures are applied to rules extracted from RBF neural networks trained on several data sets including benchmark sets from the UCI repository and a large real-world industrial data set.