Mariem Gzara

A density based algorithm for discovering clusters with varied density

Clustering is a well studied problem in data analysis and data mining. It has many areas of applications and it is used as a preprocessing step before other data mining tasks such as classification and association analysis. Discovering clusters of arbitrary shapes is a challenging task. Even though density based clustering algorithms manage to detect clusters with different shapes and sizes in large data bases with the presence of noise, they fail in handling local density variation within the data. In this paper, we propose a new algorithm based on the well known density based clustering algorithm DBSCAN. Our algorithm approximates the k nearest neighbors curve by spline interpolation and uses mathematic properties of functions to detect automatically points where the function changes concavity. Some of these points corresponds to the different levels of density within the data set. Experimental results on synthetic data sets show the efficiency of the proposed approach.

A bio-inspired hierarchical clustering algorithm with backtracking strategy

Biological entities, such as birds with their flocking behavior, ants with their social colonies, fish with their shoaling behavior and honey bees with their complex nest construction, represent a great source of inspiration in the optimization and data mining domains. Following this line of thought, we propose the Communicating Ants for Clustering with Backtracking strategy (CACB) algorithm, which is based on a dynamic and an adaptive aggregation threshold and a backtracking strategy where artificial ants are allowed to turn back in their previous aggregation decisions. The CACB algorithm is a hierarchical clustering algorithm that generates compact dendrograms since it allows the aggregation of more than two clusters at a time. Its high performance is experimentally shown through several real benchmark data sets and a content-based image retrieval system.

An ant-based algorithm for clustering

In this paper, we propose a new ant based clustering algorithm. The algorithm takes inspiration from the sound communication properties of real ants. Artificial ants communicate directly with each others in order to merge similar group of objects. The proposed algorithm was tested and evaluated. The obtained results are very encouraging in comparison with the famous k-means and some ant based clustering algorithms.

Unsupervised varied density based clustering algorithm using spline

Abstract Building upon the promising performances of density-based clustering, we present a novel density-based clustering algorithm called MDCUT (MultiDensity ClUsTering). The presented algorithm has the merit of clustering data with varied density. It operates in two phases. First, it finds the appropriate number of density levels in a data set; to do so, it uses the exponential spline mathematical process on the k-nearest neighbors’ distance. Secondly, it uses these levels as local density thresholds to determine clusters with varied densities and arbitrary shapes. We show experimentally that the MDCUT algorithm detects correctly the density levels in a data set and succeeds to discover arbitrarily shaped clusters in decreasing density order. We validate the clustering results in terms of clustering error, precision and recall rates on various data sets. MDCUT performs well in comparison to several other clustering algorithms among which the DBSCAN algorithm.

An UML class recommender system for software design

Recommendation systems provide suggestions for items that are potentially interesting for a user in a given context. The provided recommendations are extracted generally from a huge amount of data collected from several sources of information. Thus a recommendation system requires firstly a pre-treatment step to prepare the data and secondly the application of some techniques such as data mining techniques to handle and extract the knowledge to be recommended to the user from the data. Our contribution consists on proposing a Recommendation System for Software Engineering (RSSE). This system recommends UML classes in the design phase of UML classes diagrams. Our RSSE is composed by two main phases: an off-line phase in which we use a clustering algorithm to partition UML classes collected from several UML classes diagrams based on the semantic relations existing between their characteristics. We have defined a metric that measures the similarity between UML classes. The second is an online phase in which we use the obtained clusters of UML classes to propose suggestions to the user based on elements added to his UML classes diagram under construction. The proposed system is then experimentally evaluated by using a UML classes corpus collected from several UML classes diagrams. The experimental evaluation shows very encouraging ratio of useful recommendations.

FDClust: a new bio-inspired divisive clustering algorithm

Clustering with bio-inspired algorithms is emerging as an alternative to more conventional clustering techniques. In this paper, we propose a new bio-inspired divisive clustering algorithm FDClust (Artificial Fish based Divisive Clustering algorithm). FDClust takes inspiration from the social organization and the encounters of fish shoals. In this algorithm, each artificial fish (agents) is identified with one object to be clustered. Agents move randomly on the clustering environment and interact with neighboring agents in order to adjust their movement directions. Two Groups of similar objects will appear through the movement of agents in the same direction. The algorithm is tested and evaluated on several real benchmark databases. The obtained results are very interesting in comparison with Kmeans, Slink, Alink, Clink and Diana algorithms.

MDCUT 2 : a multi-density clustering algorithm with automatic detection of density variation in data with noise

Despite their adoption in many applications, density-based clustering algorithms perform inefficiently when dealing with data with varied density, imbricated and/or adjacent clusters. Clusters of lower density may be classified as outliers, and adjacent and imbricated clusters with varied density may be aggregated. To handle this inefficiency, the MDCUT algorithm (Multiple Density ClUsTering) (Louhichi et al. in Pattern Recogn Lett 93:48–57, 2017) detects multiple local density parameters to handle density variation in the data. MDCUT extracts density local levels by analyzing mathematically the interpolated k-nearest neighbors function. A clustering Sub-routine is lunched for each density level to form the clusters of that level. Compared to well-known density based clustering algorithms, MDCUT recorded good results on artificial datasets. The main drawback of MDCUT is its sensitivity to the parameter p of the used interpolation technique and the parameter k for the number of nearest neighbors. In this paper, we propose a new extension of the MDCUT algorithm to detect automatically pairs of values (ki,εi) to characterize the density levels in the data, where ki and εi stand respectively for the number of neighbors and the radius threshold for the ith density level. We study the performance of the MDCUT2 algorithm on well-known data sets by comparison to reference density based clustering algorithms. This extension has improved the previous classification results.

Clustering aggregation for improving ant based clustering

In this paper, we propose a hybridization between an antbased clustering algorithm: CAC (Communicating Ants for Clustering) algorithm [5] and a clustering aggregation algorithm: the Furthest algorithm [6]. The CAC algorithm takes inspiration from the sound communication properties of real ants. In this algorithm, artificial ants communicate directly with each other in order to achieve the clustering task. The Furthest algorithm takes as inputsm clusterings given bym different runs of the CAC algorithm, and tries to find a clustering that matches, as possible, all the clusterings given as inputs. This hybridization shows an improvement of the obtained results.

Artificial Ants for Clustering with Adaptive Aggregation Conditions: Application to Image Clustering

The world of ants is a reach source of inspiration since real ants are able to solve collectively relatively complex problems. Particularly, several ant based clustering algorithms have been proposed in the literature. These clustering models were derived from several phenomena among real ants such as cemetery organization, recognition system, building alive structures, etc. In this work, we try to adapt the properties of sound communication among real ants to resolve the clustering problem. Artificial ants move randomly on a 2D toroidal grid where objects are initially scattered at random. They communicate with each others in order to recruit ants having similar heaps of objects. We have applied this algorithm on many databases and we get very good results compared to the K-means algorithm. An application to image clustering is also realized.