Particle Swarm based Data Mining Algorithms for classification tasks

Abstract

Particle Swarm Optimisers are inherently distributed algorithms where the solution for a problem emerges from the interactions between many simple individual agents called particles. This article proposes the use of the Particle Swarm Optimiser as a new tool for Data Mining. In the first phase of our research, three different Particle Swarm Data Mining Algorithms were implemented and tested against a Genetic Algorithm and a Tree Induction Algorithm (J48). From the obtained results, Particle Swarm Optimisers proved to be a suitable candidate for classification tasks. The second phase was dedicated to improving one of the Particle Swarm optimiser variants in terms of attribute type support and temporal complexity. The data sources here used for experimental testing are commonly used and considered as a de facto standard for rule discovery algorithms reliability ranking. The results obtained in these domains seem to indicate that Particle Swarm Data Mining Algorithms are competitive, not only with other evolutionary techniques, but also with industry standard algorithms such as the J48 algorithm, and can be successfully applied to more demanding problem domains.

SwarmDock and the Use of Normal Modes in Protein-Protein Docking

“…The PSO algorithm is effective as a robust optimiser of non-linear, multi-modal functions. It has been successfully applied to many optimisation problems including neural network training, design optimisation, data mining [56], gene clustering [57] and sequence alignment [58]. More recently, PSO has been used for small molecule docking and has been shown to outperform methods based on simulated annealing and genetic algorithms [59–61].…”

Mining cancer data with discrete particle swarm optimization and rule pruning

“…In data mining, classification refers to predicting whether an instance belongs to a certain group using rules discovered from existing data via statistical techniques. Common methods include artificial neural networks, decision tree (DT), support vector machine (SVM), genetic algorithm (GA), and more recently an evolutionary algorithm named particle swarm optimization (PSO), of which PSO has demonstrated to competitive in classification tasks, notably in cancer identification and prediction [7]. Based on the standard PSO, Yeh [8] proposed an algorithm named discrete particle swarm optimization (DPSO), which obtains a high accuracy in breast cancer classification while being more efficient than the original PSO, and make it a suitable algorithm to be used in cancer detection applications.…”

“…Rule pruning address the issue of over fitting the training data by removing the irrelevant terms from the rule, and improves the predictive power of the rule, and in the meantime simplifies it [9] [10]. While in conventional pruning procedure, one attribute is taken out at a time to examine the rule quality [7][10], for rule which there are multiple limitation conditions in one attribute, the influence of individual parameter inside each attribute is overlooked, and thus it is worth examine each parameter separately.…”

Applying Particle Swarm Optimization algorithm to roundness measurement

“…The authors combined the concept of swap operator, swap sequence, particle swarm optimization, and redefined some operators to resolve the TSP. Sousa et al established an algorithm for data mining on the basis of particle swarm optimization [13]. Pang et al [14] incorporated a modified PSO, which constructed a mapping from continuous space to discrete permutation space, with local search technique to solve TSP.…”