Ruhaizan Ismail

An Agent Based Rough Classifier for Data Mining

This paper proposes a new agent based approach in rough set classification theory. Rough set is one of data mining techniques for classification. It generates rules from large database and it has mechanism to handle noise and uncertainty in data. However, to produce a rough classification model or rough classifier is highly computational especially in its reduct computation phase which is an np-hard problem. These have contributed to the generation of large amount of rules and lengthy processing time. To resolve the problem, an agent based algorithm is embedded within the rough modelling framework. In this study, the classifier are based on creating agent within the main modelling processes such as reduct computation, rules generation and attribute projections. Four main agents are introduced i.e. interaction agent, weighted agent, reduction agent and default agent. The experimental result shows that the proposed method reduces the running time with a comparative classification accuracy and number of rules.

An agent model for rough classifiers

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.

Associative prediction model and clustering for product forecast data

Association rules are adopted to discover the interesting relationship and knowledge in a large dataset. Knowledge may appear in terms of a frequent pattern discovered in a large number of production data. This knowledge can improve or solve production problems to achieve low cost production. To obtain knowledge and quality information, data mining can be applied to the manufacturing industry. In this study, we used one of the association rule approach, i.e. Apriori algorithm to build an associative prediction model for product forecast data. Also, we adopt the simplest method in clustering, k-means algorithm to attain the link between patterns. The real industrial product forecast data for one year duration is used in the experiment. This data consists of 42 products with two important attributes, i.e. time in the week and required quantity. Since the data mining processes need a large amount of data, we simulated these data by using the Monte Carlo technique to obtain another 15 years of simulated forecast data. There are two main experiments for the association rules mining and clustering. As a result, we obtain an associative prediction model and clustering for the forecasting data. The extracted model provides the prediction knowledge about the range of production in a certain period.

Using rough set theory for mining the level of hearing loss diagnosis knowledge

This paper focused on the development of diagnosis knowledge model of the level of hearing loss in the audiology clinic patients using rough set theory. A knowledge model contains a set of knowledge via rules that are obtained from mining certain amount of data. These data consist of valuable knowledge that impossible for the audiologist or audio therapist to extract without powerful mining techniques or tools. These rules help doctors in decision making such as setting up new strategy to improve the efficiency of the operation. In this work, a data mining technique, rough set theory was used for the knowledge modelling. It was used based on its capability of handling uncertain data that often occurs in real world problems. The results from the modelling produced a classifier called rough classifier. The classifier was used to classify the level of hearing loss. A total of 500 data obtained from the audiology clinic. The data consisted of 24 attributes from four categories namely demography, antenatal, neonatal and medical categories. These attributes were used as an input and one attribute called diagnostic category as an output. In order to facilitate the modelling process requirement, these attributes have been gone a pre-process stage. The best model has been obtained from 10 experiments using 10 sets of different training and test data. The experiment showed promising results with 76% accuracy. The developed knowledge model has a great potential to be embedded in the development of the medical decision support system.

Data mining in production planning and scheduling: A review

The paper reviews about the data mining tasks and methods, and its application in production planning and scheduling. Data mining will be reviewed in four classifications of data mining systems according to the kinds of databases mined, knowledge to discover, techniques utilized and the applications adapted. This paper also reviews in production planning and scheduling that focused in time frame range either short- to mid-range or long-range planning. In production planning, there are a lot of planning such as process planning, strategic capacity planning, aggregate planning, master scheduling, material requirements planning and order scheduling. From these activities different problems are arise because of the different time, product and environment of production.

Agent based data classification approach for data mining

Classification is one of the tasks in data mining. The form of classifier depends on the classification technique used. For example, neural network produce a set of weight as a classifier, regression form an equation as a predictor while decision tree, C4.5, CART, Rough Set and Bayesian theory generate set of rules known as rule based classifier. Rules are more interpretable by human when compared to other form of classifiers. The process of classification involves applying the rules onto a set of unseen data. There are many issues appeared in rule application process such as more than one rule match, multiple scanning of large rule base and uncertainty. In this study an agent based approach is proposed to improve the rule application process. The proposed agents are embedded within the standard rule application techniques. The result shows the significant improvements in classification time and the number of matched rules with comparable classification accuracy.

Development of knowledge model for insurance product decision using the associative classification approach

Individual protection, physically or mentally, is very important for someone living in a risk environment. Insurance is one of the individual protections due to accident, blaze, critical diseases or death. Insurance company plays a critical role in providing competitive product insurance that covers flexible features depend on customer requirements. In order to compete with other competitors and fulfill the customer needs, the company needs a wise and proper business strategy. The insurance company needs extra knowledge on the potential customer whom can give a positive response to the insurance product being offered. In this paper we proposed an associative classification model to develop a knowledge model for determining the best class solution for insurance policy dataset. We enhanced the classification-based association of associative classification by using a heuristic to process two types of decision rules. The decision rules types were the correctly classified rules and the verified uncertain classified rules. The finding showed that the type of products could be proposed in a new insurance policy based on individual profiles.

Managing the air cargo offload problems using rough set theory

This paper focuses on the development of knowledge model for a prediction of air cargo offload. A knowledge model is a model containing a set of knowledge via rules that has been obtained from mining certain amount of data. These rules might help the management in major decision making such as setting up new strategy. In this study, an intelligent technique for data mining called a rough set theory was used for the knowledge modelling. Rough set technique has been used based on its capability of handling uncertain data often occurs in the real world problem. As a result, a rough classifier was produced and has been used for offload prediction in four travelling sectors. A total of 267 data were obtained from a Malaysian air cargo company. There were eight attributes used as input and one attribute as an output. Data has been through a pre-processing stage to facilitate the requirement of the modelling process. A total of ten experiments using ten sets of different data have been conducted. The best model was selected from the total models generated from the experiment. The model has given a promising result with 100% accuracy. The rules obtained have contributing to offload problem knowledge but need further investigation for more comprehensive knowledge decision model.