Clustering, the unsupervised classification of objects into groups, is a widely used technique in exploratory data analysis. The clustering problem is a very complex one, and a popular heuristic for solving it is the Simulated Annealing (SA) algorithm. SA is an approximation algorithm that involves generating a neighborhood solution by perturbing the current solution in a small, yet meaningful way. This new solution is accepted with a probability of 1 if it is quantitatively better than the current solution, and accepted according to the Metropolis criterion otherwise. Cluster quality is measured using the Sum of Squared Error (SSE) criterion. This paper presents an SA algorithm that uses a new type of perturbation to generate solutions. Whereas most SA clustering algorithms perturb data point memberships directly, our algorithm perturbs a randomly chosen center using Gaussian mutation, and then reassigns data points in a nearest neighbor fashion. Experimental results on a diverse collection of data sets demonstrate that our algorithm has comparable effectiveness to other SA algorithms, while being much faster due to its simplicity.

Classified ad sites routinely process hundreds of thousands to millions of posted ads, and only a small percentage of those may be fraudulent. Online scammers often go through a great amount of effort to make their listings look legitimate. Examples include copying existing advertisements from other services, tunneling through local proxies, and even paying for extra services using stolen account information. This paper focuses on applying knowledge discovery concepts towards the detection of online, classified fraud. Traditional data mining is used to extract relevant attributes from an online classified advertisements database and machine learning algorithms are applied to discover patterns and relationships of fraudulent activity. With our proposed approach, we will demonstrate the effectiveness of applying data mining techniques towards the detection of fraud in online classified advertisements.

In the present paper we propose an unsupervised attribute ranking method based on evaluating the quality of clustering that each attribute produces by partitioning the data into subsets according to its values. We use the Minimum Description Length (MDL) principle to evaluate the quality of clustering and describe an algorithm for attribute ranking and a related clustering algorithm. Both algorithms are empirically evaluated on benchmark data sets. The experiments show that the MDL-based ranking performs closely to the supervised information gain ranking and thus improves the performance of the EM and k-means clustering algorithms in purely unsupervised setting.

We propose a new method for the morphological galaxy classification which incorporates two main contributions: (i) the generation of artificial images of galaxies through geometric transformations to be used as additional examples in the training phase, (ii) the use of a novel hierarchical classifier for hierarchical galaxy classification. An additional classifier distinguishes galaxies from stars based on geometrical moments. The proposed method was tested with two different astronomical databases. The results found show that the hierarchical classification method has a higher performance than flat classification, and that the use of artificial examples and oversampling provide a significant improvement in performance.

K-means is undoubtedly the most widely used partitional clustering algorithm. Unfortunately, the nearest neighbor search step of this algorithm can be computationally expensive, as the distance between each input vector and all cluster centers need to be calculated. To accelerate this step, a computationally inexpensive distance estimation method can be tried first, resulting in the rejection of candidate centers that cannot possibly be the nearest center to the input vector under consideration. This way, the computational requirements of the search can be reduced as most of the full distance computations become unnecessary. In this paper, a fast nearest neighbor search method that rejects impossible centers to accelerate the k-means clustering algorithm is presented. Our method uses geometrical relations among the input vectors and the cluster centers to reject many unlikely centers that are not typically rejected by similar approaches. Experimental results show that the method can reduce the number of distance computations significantly without degrading the clustering accuracy.

A very promising tool for data mining and bioinformatics is ensemble gene (feature) selection. Ensemble feature selection is the process of performing multiple runs of feature selection and then aggregating the results into a final ranked list. However, a central question of ensemble feature selection is how to aggregate the individual results into a single ranked feature list. There are a number of techniques available, ranging from simple to complex; the question is which one to choose. This paper is a comprehensive study on the use of nine different rank aggregation techniques for building classification models to use gene microarray data for distinguishing between cancerous and non-cancerous cells (or between patients who did or did not respond well to cancer treatment). The techniques are tested using an ensemble with twenty-five feature selection techniques and fifty iterations along with eleven bioinformatics datasets and five learners. Our results show that Lowest Rank is the worst performing aggregation technique by a clear margin. The other techniques perform similarly well and a simple technique (e.g., Mean aggregation) is preferable due to computation time and the limited possible benefit of a more complex technique. To our knowledge there has never been a study this intensive on the classification abilities of rank aggregation techniques in the field of bioinformatics.

The state-of-the-art Data-Peeler algorithm extracts closed patterns in n-ary relations. Because it refines both a lower and an upper bound of the pattern space, Data-Peeler can, in some circumstances, guarantee that a region of that space does not contain any closed n-set satisfying some relevance constraint. Whenever it happens, such region is not explored and some time is saved. This paper shows that some constraints, which Data-Peeler can efficiently enforce, define useful patterns in the context of a relation with groups of elements in arbitrary dimensions. For instance, it can list the so-called straddling biclusters, which cover at least some given portions of every group. It can discover, as well, closed n-sets that discriminate a group from the others. The experimental section focuses on the latter case. It shows that Data-Peeler is highly competitive despite its general enumeration principles and its expressive class of constraints that open up new applicative perspectives.

The paper deals with the application of the Case-Based Reasoning Paradigm (CBR) in Design Support Systems in Architecture. Based on the finding that promising concepts and systems do exist in architecture the question as to why they do not gain the anticipated success is explored. In search for reasons a comprehensive comparison between the cognitive model and the derived conceptual method, theoretical contemplations of architectural design as well as the actual application of the method in CBR systems in Architecture, manifests the core of the work presented.

We investigate a method for applying CBR to a source of data where there are no negative exemplars. Our problem domain is one of recommending characteristics of multidisciplinary collaborators based on a collection of funded grants. Thus, there are no negative exemplars. Lacking sufficient domain knowledge, we seek to apply a feedback algorithm to learn weights even in the absence of negative exemplars. Our approach is based on the assumption that well aligned cases, cases where similar problems have similar solutions, are better suited for learning feature weights. Our approach clusters the problem and solution spaces separately to identify well aligned cases. We also identify poorly aligned cases that may hinder effective learning of weights, and exclude them. The clusters of well aligned cases provide a means to utilize feedback algorithms. We use two methods, case alignment and case cohesion, to show that our approach succeeds in identifying well aligned cases. We also compare our approach to a method based on single class learning, a machine learning approach for reasoning without negatives. Our results show that our approach is viable to learning weight in the absence of negative exemplars.

The success of the Case Based Reasoning (CBR) system depends on the quality of the case data. This quality is dedicated to the study of the case base competence which is measured by the range of problems that can be satisfactorily solved. In fact, modeling case-base competence is a clamorous issue in the discipline of CBR. However, the existence of erroneous cases as noises and the non uniform problem distributions has not been considered in the proposed computing competence. In this paper, we proposea novel case base competence model based on Mahalanobis distance and a clustering technique named DBSCAN-GM. The advantage of this newly proposed model is its high accuracy for predictingcompetence. In addition, it is not sensitive to noisy cases and it takes account the situation of the distributed case-base.Withal, we contest that this model has aconspicuous role to play in future CBR research infields such as the development of new policies for maintainingthe case base.