1. UKM Learning and Research Repository
2. Master Thesis / Tesis Master
3. Faculty of Information Science and Technology / Fakulti Teknologi dan Sains Maklumat

Title:	Particle swarm optimization algorithm for anomalous association rules mining
Authors:	Vahid Beiranvand (P59318)
Supervisor:	Azuraliza Abu Bakar, Prof. Dr.
Keywords:	Rules mining Particle Swarm Optimization (PSO) Algorithm Universiti Kebangsaan Malaysia -- Dissertations Dissertations, Academic -- Malaysia
Issue Date:	8-Oct-2012
Description:	An anomaly is an observation that does not conform to the expected normal behaviour. Anomaly detection can provide useful information in a variety of real-world problems such as fraud detection, fault detection, and outbreak detection. One of the potential methods for anomaly detection is Association Rule Mining (ARM). Anomalous rule mining is a variation of ARM dealing with unusual knowledge that might be of interest for the user and has its potential use for anomaly detection. One of the important tasks of ARM is determining the appropriate minimum threshold values for support and confidence factors. A large value of minimum threshold will limit the generation of important rules whereas a small minimum threshold value will generate a huge amount of less important rules. Both factors are often determined by the expert user or by trial-and-error, where the features of the dataset are not considered effectively. Therefore, to handle this problem, this study aims to propose a PSO algorithm to optimise the minimum threshold value for the support and confidence factors in anomalous association rules. It attempts to improve the efficiency of the anomalous ARM process by providing suitable threshold values for support and confidence. The proposed approach consists of two main steps. First, the Particle Swarm Optimization (PSO) algorithm is developed to determine both the minimum support and minimum confidence of anomalous rules as the preprocessing phase. Secondly, the obtained minimum support and confidence are employed to extract anomalous rules. Four benchmark datasets are used to evaluate the performance of the algorithm. It is measured in terms of the number of extracted anomalous rules compared to the number of regular association rules. The results showed that the proposed algorithm generates a less number of anomalous rules with high confidence values compared to the previous approach. The study showed that the proposed algorithm is able to obtain the appropriate values for minimum support and confidence leading to extraction of a manageable number of high confidence anomalous rules.,Certification of Master's / Doctoral Thesis" is not available
Pages:	93
Call Number:	Q337.3.B435 2012 3tesis
Publisher:	UKM, Bangi
Appears in Collections:	Faculty of Information Science and Technology / Fakulti Teknologi dan Sains Maklumat

Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.