Marco Comuzzi

Combining Apriori heuristic and bio-inspired algorithms for solving the frequent itemsets mining problem

Abstract Exact approaches to Frequent Itemsets Mining (FIM) are characterised by poor runtime performance when dealing with large database instances. Several FIM bio-inspired approaches have been proposed to overcome this issue. These are considerably more efficient from the point of view of runtime performance, but they still yield poor quality solutions. The quality of the solution, i.e., the number of frequent itemsets discovered, can be increased by improving the randomised search of the solutions space considering intrinsic features of the FIM problem. This paper proposes a new framework for FIM bio-inspired approaches that considers the recursive property of frequent itemsets, i.e., the same feature exploited by the Apriori exact heuristic, in the search of the solution space. We define two new approaches to FIM, namely GA-Apriori and PSO-Apriori, based on the proposed framework, which use genetic algorithms and particle swarm optimisation, respectively. Extensive experiments on synthetic and real database instances show that the proposed approaches outperform other bio-inspired ones in terms of runtime performance. The results also reveal that the performance of PSO-Apriori is comparable to the one of exact approaches Apriori and FPGrowth in respect of the quality of solutions found. We also show that PSO-Apriori outperforms the recently developed BATFIM algorithm when dealing with very large database instances.

SS-FIM: Single Scan for Frequent Itemsets Mining in Transactional Databases

The quest for frequent itemsets in a transactional database is explored in this paper, for the purpose of extracting hidden patterns from the database. Two major limitations of the Apriori algorithm are tackled, (i) the scan of the entire database at each pass to calculate the support of all generated itemsets, and (ii) its high sensitivity to variations of the minimum support threshold defined by the user. To deal with these limitations, a novel approach is proposed in this paper. The proposed approach, called Single Scan Frequent Itemsets Mining (SS-FIM), requires a single scan of the transactional database to extract the frequent itemsets. It has a unique feature to allow the generation of a fixed number of candidate itemsets, independently from the minimum support threshold, which intuitively allows to reduce the cost in terms of runtime for large databases. SS-FIM is compared with Apriori using several standard databases. The results confirm the scalability of SS-FIM and clearly show its superiority compared to Apriori for medium and large databases.

Impact analysis of ERP post-implementation modifications

An innovative framework to support identification and assessment of ERP post-implementation modifications.Identify impact of proposed modifications based on a meta-model of ERP entity dependencies and mechanisms for impact propagation on the design time structure and run time business operations.A software tool is proposed that implements the framework and is instantiated for two pseudo-real ERP installations.Framework tested with ERP experts to assess fit for purpose and user satisfaction. ERP systems evolve in the post-implementation phase because of changing business requirements. Post-implementation changes are likely to decrease the quality of ERP systems and of the data that they use, which negatively impacts organisational performance. We propose a framework for impact analysis of ERP post-implementation modifications. Our framework allows mapping dependencies among ERP system components and, based on these dependencies, automatically assessing the impact of a proposed change on both the design-time structure and run-time landscape of the system through a novel set of impact metrics. The framework also provides semi-automatic support to safely terminating the running process instances affected by change. The framework is evaluated with expert users in two pseudo-real ERP system implementations.

GPU-based Bio-inspired Model for Solving Association Rules Mining Problem

We explore in this paper the application of bioinspired approaches to the association rules mining (ARM) problem for the purpose of accelerating the process of extracting the correlations between items in sizeable data instances. A new bio-inspired GPU-based model is proposed, which benefits from the massively GPU threading by evaluating multiple rules in parallel on GPU. To validate the proposed model, the most used bio-inspired approaches (GA, PSO, and BSO) have been executed on GPU to solve wellknown 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 the genetic algorithm outperforms PSO and BSO. Moreover, it outperforms the state-of-the-art GPUbased ARM approaches when dealing with the challenging Webdocs instance.

GA-Apriori: Combining Apriori Heuristic and Genetic Algorithms for Solving the Frequent Itemsets Mining Problem

Finding frequent itemsets is a popular data mining problem, aiming to extract hidden patterns from a transactional database. Several bio-inspired approaches to solve this problem have been proposed to overcome the poor performance of exact algorithms, such as Apriori and FPGrowth. Approaches based on genetic algorithms are among the most efficient ones from the point of view of runtime performance, but they are still inefficient in terms of solution’s quality, i.e., the number of frequent itemsets discovered. To deal with this issue, we propose in this paper a new genetic algorithm for finding frequent itemsets called GA-Apriori, in which the crossover and mutation operators are defined by taking into account the Apriori heuristic principle. The results of our evaluation show that GA-Apriori outperforms other approaches to frequent itemset mining based on genetic algorithms, especially when dealing with large instances. The experiments also show that GA-Apriori is competitive with exact approaches in terms of the number of frequent itemsets discovered.

Diversification Heuristics in Bees Swarm Optimization for Association Rules Mining

Association rules mining is becoming more challenging with the large transactional databases typical of modern times. Conventional exact algorithms for association rules mining struggle to cope with very large databases, especially in terms of run-time performance. To address this problem, several evolutionary and swarm intelligence-based approaches have been proposed. One of these is HBSO-TS, which is a hybrid approach combining Bees Swarm Optimization with Tabu Search and has been shown to outperform other state-of-the art bio-inspired approaches. The main drawback of HBSO-TS is that while the intensification is improved using Tabu Search, the diversification remains unchanged compared to BSO-ARM, i.e., the first approach proposed in the literature using Bees Swarm Optimization for association rules mining. To ensure a better balance between intensification and diversification, this paper proposes two new heuristics for determining the search area of the bees. We conducted experimental evaluation on well known data instances to show that both heuristics improve the performance of HBSO-TS. Moreover, we show the usefulness of our heuristics in the special case of mining association rules from diversified data, as in the case of Weblog mining.

Optimal Paths in Business Processes: Framework and Applications

We present an innovative framework for calculating optimal execution paths in business processes using the abstraction of workflow hypergraphs. We assume that information about the utility associated with the execution of activities in a process is available. Using the workflow hypergraph abstraction, finding a utility maximising path in a process becomes a generalised shortest hyperpath problem, which is NP-hard. We propose a solution that uses ant-colony optimisation customised to the case of hypergraph traversal. We discuss three possible applications of the proposed framework: process navigation, process simulation, and process analysis. We also present a brief performance evaluation of our solution and an example application.

Optimal directed hypergraph traversal with ant-colony optimisation

Abstract Directed hypergraphs are an extension of directed graphs in which edges connect a set of source nodes to a set of target nodes. Unlike graphs, they can capture complex relations in network structures that go beyond the union of pairwise associations. They are widely applied in a variety of different domains, such as finding pathways in chemical reaction networks or minimising propositional Horn formulas. Calculating optimal paths in hypergraphs in the general case is an NP-hard problem, which can be solved in polynomial time only when utility functions hold specific properties. We present in this paper an approach to search for optimal hypergraph paths in the general case based on ant colony optimisation. Ant colony optimisation is an evolutionary meta-heuristic that is particularly suitable to combinatorial problems, such as optimal graph traversal. We present an experimental evaluation using artificially-generated hypergraphs and discuss innovative applications of the proposed approach in the domains of industrial engineering and chemical informatics.

Towards a Design Space for Blockchain-Based System Reengineering

We discuss our ongoing effort in designing a methodology for blockchain-based system reengineering. In particular, we focus in this paper on defining the design space, i.e., the set of options available to designers when applying blockchain to reengineer an existing system. In doing so, we use a practice-driven approach, in which this design space is constructed bottom-up from analysis of existing blockchain use cases and hands-on experience in real world design case studies. Two case studies are presented: using blockchain to reengineer the meat trade supply chain in Mongolia and blockchain-based management of ERP post-implementation modifications.

Fuzzy analytic network process for evaluating ERP post-implementation alternatives

Because of the unreliability of human experts judgements, fuzzy systems are widely applied in decision making problems related to adoption, implementation and maintenance of ERP systems. This paper presents a novel application of the fuzzy Analytical Network Process (ANP) for evaluating ERP post-implementation alternatives. The proposed framework aims at ranking different alternatives to implement a given post-implementation business requirement based on experts perception of implementation effort and risk. Regarding decision criteria, the ANP network considers two levels. At the higher level of abstraction we consider technical effort, organisational effort and long-term risk, whereas at the lower level we consider different strategies for implementing given business objects, functions and processes in an existing ERP system. Decision makers preferences are translated into preference weights using triangular fuzzy numbers. An example is presented to show the application of the proposed framework.