Youcef Djenouri

GPU-based bees swarm optimization for association rules mining

Association rules mining (ARM) is a well-known combinatorial optimization problem aiming at extracting relevant rules from given large-scale datasets. According to the state of the art, the bio-inspired methods proved their efficiency by generating acceptable solutions in a reasonable time when dealing with small and medium size instances. Unfortunately, to cope with large instances such as the webdocs benchmark, these methods require more and more powerful processors and are time expensive. Nowadays, computing power is no longer a real issue. It can be provided by the power of emerging technologies such as graphics processing units (GPUs) that are massively multi-threaded processors. In this paper, we investigate the use of GPUs to speed up the computation. We propose two GPU-based bees swarm algorithms for association rules mining (single evaluation in GPU, SE-GPU and multiple evaluation in GPU, ME-GPU). SE-GPU aims at evaluating one rule at a time where each thread is associated with one transaction, whereas ME-GPU evaluates multiple rules in parallel on GPU where each thread is associated with several transactions. To validate our approaches, the two algorithms have been executed to solve well-known large ARM instances. Real experiments have been carried out on an Intel Xeon 64 bit quad-core processor E5520 coupled to an Nvidia Tesla C2075 GPU device. The results show that our approaches improve the execution time up to 100

Data Mining-Based Decomposition for Solving the MAXSAT Problem: Toward a New Approach

\times

Multilevel Clustering of Induction Rules for Web Meta-knowledge

× over the sequential mono-core bees swarm optimization-ARM algorithm. Moreover, the proposed approaches have been compared with CPU multi-core ones (1–8 cores). The results show that they are faster than the multi-core versions whatever is the number of used cores.

Parallel BSO Algorithm for Association Rules Mining Using Master/Worker Paradigm

Abstract Business process analysis is a key activity that aims at increasing the efficiency of business operations. In recent years, several data mining based methods have been designed for discovering interesting patterns in event logs. A popular type of methods consists of applying frequent itemset mining to extract patterns indicating how resources and activities are frequently used. Although these methods are useful, they have two important limitations. First, these methods are designed to be applied to original event logs. Because these methods do not consider other perspectives on the data that could be obtained by applying data transformations, many patterns are missed that may represent important information for businesses. Second, these methods can generate a large number of patterns since they only consider the minimum support as constraint to select patterns. But analyzing a large number of patterns is time-consuming for users, and many irrelevant patterns may be found. To address these issues, this paper presents an improved event log analysis approach named AllMining. It includes a novel pre-processing method to construct multiple types of transaction databases from a same original event log using transformations. This allows to extract many new useful types of patterns from event logs with frequent itemset mining techniques. To address the second issue, a pruning strategy is further developed based on a novel concept of pattern coverage, to present a small set of patterns that covers many events to decision makers. Results of experiments on real-life event logs show that the proposed approach is promising compared to existing frequent itemset mining approaches and state-of-the-art process model algorithms.

Fast and effective cluster-based information retrieval using frequent closed itemsets

This article explores advances in the data mining arena to solve the fundamental MAXSAT problem. In the proposed approach, the MAXSAT instance is first decomposed and clustered by using data mining decomposition techniques, then every cluster resulting from the decomposition is separately solved to construct a partial solution. All partial solutions are merged into a global one, while managing possible conflicting variables due to separate resolutions. The proposed approach has been numerically evaluated on DIMACS instances and some hard Uniform-Random-3-SAT instances, and compared to state-of-the-art decomposition based algorithms. The results show that the proposed approach considerably improves the success rate, with a competitive computation time that’s very close to that of the compared solutions.

Intelligent mapping between GPU and cluster computing for discovering big association rules

The association rules mining (ARM) problem is one of the most important problems in the area of data mining. It aims at finding all relevant association rules from transactional databases. It is CPU time intensive and requires a huge computing power when dealing with large transactional databases. To deal with this issue, Graphics Processing Units (GPUs) are a powerful tool to speed up the search process. However, their performance is closely subject to thread/branch divergence resulting from the single instruction multiple data parallel model of GPUs. In this paper, we propose three approaches based on database reorganization, aiming to reduce thread divergence in GPU‐based bees swarm optimization metaheuristic for ARM, respectively, named block‐based reordering, transactions‐based reordering, and transactions‐based reordering with median value. Theoretical and experimental studies have been carried out using well‐known large ARM instances. The experiments have been performed on an Intel Xeon 64 bit quad‐core processor E5520 coupled to Nvidia Tesla C2075 448 cores. The results show that the proposed approaches minimize considerably the number of thread divergence and improve the overall execution time. Indeed, the number of thread divergence occurrences has been reduced by up to eight times making the execution much faster. Copyright

How to exploit high performance computing in population-based metaheuristics for solving association rule mining problem

The current World Wide Web is featured by a huge mass of knowledge, making it difficult to exploit. One possible way to cope with this issue is to proceed to knowledge mining in a way that we could control its volume and hence make it manageable. This paper explores meta-knowledge discovery and in particular focuses on clustering induction rules for large knowledge sets. Such knowledge representation is considered for its expressive power and hence its wide use. Adapted data mining is proposed to extract meta-knowledge taking into account the knowledge representation which is more complex than simple data. Besides, a new clustering approach based on multilevel paradigm and called multilevel clustering is developed for the purpose of treating large scale knowledge sets. The approach invokes the k-means algorithm to cluster induction rules using new designed similarity measures. The developed algorithms have been implemented on four public benchmarks to test the effectiveness of the multilevel clustering approach. The numerical results have been compared to those of the simple k-means algorithm. As foreseeable, the multilevel clustering outperforms clearly the basic k-means on both the execution time and success rate that remains constant to 100 % while increasing the number of induction rules.

Maintenance algorithm for high average-utility itemsets with transaction deletion

The extraction of association rules from large transactional databases is considered in the paper using cluster architecture parallel computing. Motivated by both the successful sequential BSO-ARM algorithm, and the strong matching between this algorithm and the structure of the cluster architectures, we present in this paper a new parallel ARM algorithm that we call MW-BSO-ARM for master/worker version of BSO-ARM. The goal is to deal with large databases by minimizing the communication and synchronization costs, which represent the main challenges that faces any cluster architecture. The experimental results are very promising and show clear improvement that reaches \(300\,\%\) for large instances. For examples, in big transactional database such as WebDocs, the proposed approach generates \(10^{7}\) satisfied rules in only 22 min, while a previous GPU-based approach cannot generate more than \(10^{3}\) satisfied rules into 10 h. The results also reveal that MW-BSO-ARM outperforms the PGARM cluster-based approach in terms of computation time.