Elsevier

Applied Soft Computing

Volume 11, Issue 2, March 2011, Pages 2239-2245
Applied Soft Computing

An agent model for rough classifiers

https://doi.org/10.1016/j.asoc.2010.08.004Get rights and content

Abstract

This paper proposes a new agent-based approach in rough set classification theory. In data mining, the rough set technique is one classification technique. It generates rules from a large database and has mechanisms to handle noise and uncertainty in data. However, producing a rough classification model or rough classifier is computationally expensive, especially in its reduct computation phase: this is an NP-hard problem. These problems have brought about the generation of large amount of rules and high processing time. We solve these problems by embedding an agent-based algorithm within the rough modelling framework. In this study, the classifiers are based on creating agents within the main modelling processes such as reduct computation, rules generation and attribute projections. Four main agents are introduced: the interaction agent, weighted agent, reduction agent and default agent. We propose a heuristic for the default agent to control its searching activity. Experiments show that the proposed method significantly reduces the running time and the number of rules while maintaining the same classification accuracy.

Introduction

Data mining is an Artificial Intelligence (AI) technology that comprises various techniques to extract comprehensible, hidden and useful information from a corpus of data. AI technology makes it possible to discover hidden trends and patterns in large amounts of data. The output of a data mining exercise can take the form of patterns, trends or rules that are implicit in the data. The rough set is a technique to identify and recognize common patterns in data, especially in uncertain and incomplete data. The Polish logician Zdzislaw Pawlak introduced this concept in the early 1980s [1]. It is currently a popular technique in AI. The mathematical foundations of this method are based on the set approximation of the classification space [1]. The rough set philosophy is founded on the assumption that we can associate some information such as data or knowledge with the universe of discourse. The rough set technique for data mining includes processes such as clustering data with indiscernibility relations to form equivalence classes, computing reducts with the concept of discernibility, and generating rules by extracting knowledge.

These processes are complex, computationally expensive, and involve many optimization problems, especially in reduct computation and rules generation. For example, using rough sets in mining a dataset of size 500 or less, more than 100,000 rules will be generated. The process of selecting the relevant rules is crucial, as it increases the quality of the knowledge model and therefore the quality of decision-making. Several researchers have discussed agent technology as a possible solution to various computational and optimization problems in rough set modelling for data mining [2], [3], [4], [5].

An agent is a program that can assist or act on behalf of users to do multiple tasks or repetitive tasks [6]. Agents should be able to learn from experience and to act autonomously to handle ever-changing tasks. Embedding agent technology in data mining techniques, particularly the rough set technique, has good potential to solve the computational and optimization problems. In this study, we propose an agent-based rough classifier based on creating agents within the main modelling processes such as reduct computation, rules generation and attribute projections. The purpose of this paper is to illustrate the use of agents within rough classification theory, in order to improve mining speed and maintain the quality of knowledge.

The paper is organized as follows. Section 2 provides a basic idea of the rough classification algorithm used in this study. Section 3 details the proposed method and the use of agents in rough classifiers. It describes the concept and type of agent that will be used and presents the new algorithm. Section 4 presents describes the experimental results, and Section 5 presents the conclusion.

Section snippets

Preliminaries of rough classifiers

The theory of rough sets is concerned with the analysis and modelling of classification and decision problems involving vagueness, imprecision, or uncertain or incomplete information. It provides techniques to analyze data dependencies, to identify fundamental factors and to discover deterministic or non-deterministic rules from data [7], [9], [10]. An Information System (IS) is the most basic kind of knowledge. It consists of a set of tuples, in which each tuple is a collection of attribute

Proposed Rough Classifier Agent (RCA)

Our method, the Rough Classifier Agent (RCA), uses agents to handle several points of complexity in rough classification modelling. The main difference between RCA and previous agent-based rough set techniques is that it situates agents within the default rules generation framework [7], [8], [9], [10]. RCA introduces four main agents: the weighted agent, interaction agent, reduction agent and default agent. Fig. 2 shows the proposed agent in rough classification modelling, modifying the

Experiments

The proposed RCA improves the Emergent Based Approach (EBA) introduced by Hassan and Tazaki [3] by reducing its running time significantly while maintaining the number of rules and classification accuracy. The RCA used in the default rule generation framework controls the number of iterations and excludes unnecessary traversals of the mining path. Eight datasets from the UCI machine learning data repository are used for the experiment: the Cleveland Heart Disease, Breast Cancer, Australian

Conclusions

The experiments show that our agent-based technique (RCA) has reduced running time significantly while maintaining the accuracy of classification and the number of rules. The second experiment showed that the agent can reduce the size of the dataset through reduction while maintaining the quality of knowledge. This can be seen through the lower number of rules generated with high accuracy. This criterion is important in agent-based data mining in finding a good knowledge model from a large and

References (12)

  • Z. Pawlak

    Rough Set: Theoretical Aspect of Reasoning about Data

    (1991)
  • Y. Hassan et al.

    Interpretation of rough neural networks as emergent model

  • Y. Hassan et al.

    Emergent rough set data analysis

  • A. Skowron

    Approximate reasoning by agents

  • X.T. Hu et al.

    A new rough sets model based on database systems

  • Soft Computing and Personalized Information Provision,...
There are more references available in the full text version of this article.

Cited by (5)

  • An agent-based intelligent algorithm for uniform machine scheduling to minimize total completion time

    2014, Applied Soft Computing Journal
    Citation Excerpt :

    After the experiment is broken down into a series of experiments, genetic algorithms are then used to analyze each factor which then becomes a multi-objective optimization study [22]. Bakar et al. [23] proposed a new agent-based approach in rough set classification theory. The classifiers are based on creating agents within the main modeling processes such as reduct computation, rules generation and attribute projections.

  • Research of multi-Agent cognition and decision

    2017, ACM International Conference Proceeding Series
  • An agent model for Incremental Rough Set-based Rule Induction: A big data analysis in sales promotion

    2013, Proceedings of the Annual Hawaii International Conference on System Sciences
  • An agent model for incremental rough set-based rule induction in customer relationship management

    2012, Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
View full text