Lizawati Mi Yusuf

Features Discovery for Web Classification Using Support Vector Machine

The ever fast-expanding web information resources pose a big challenge to internet users seeking the most relevant, latest and quality information. The sheer vast amount of web information has resulted in restructuring of the resources. Thus, an appropriate web classification method needs to be established in order for quality web information to be accessed. This paper intends to discuss the web document features that classify the web information resources. Six web document features have been identified which are text, meta tag and title (A), title and text (B), title (C), meta tag and title (D), meta tag (E) and text (F). The Support Vector Machine (SVM) method is used to classify the web document while four types of kernels namely: Radial Basis Function (RBF), linear, polynomial and sigmoid kernels was applied to test the accuracy of the classification. The studies show that the text, meta tag and title (A) features is the best features for classification of web document that employs the four kernels followed by the features on title and text (B) as well as the features on meta tag and title (C). The studies also found that the linear kernel is the best kernel in classifying the web document compared to the RBF, polynomial and sigmoid kernel.

The parallel AGE variances method for temperature prediction on laser glass interaction

This paper describes the formula of three-dimensional parabolic equations for cylindrical coordinate glass that is used for mathematical simulation in simulating the temperature behavior of the laser glass cutting. There are three methods have been used for the simulation purposes which is the Alternating Group Explicit (AGE) which compromises two variances that is Brian and Douglas variant, and Gauss Seidel Red Black method. The simulation for these three methods is conducted in a parallel computing environment in order to speed up the calculation process and to achieve an accurate and convergence results. From the simulation, the results will be compared by conducting parallel performance measurement, which include execution time, speedup, efficiency, effectiveness and temporal performance.

The architecture of scientific computing library portal on sequential algorithm based on partial differential equation

Scientific Computing Library Portal (SCLP) is a system for scientific researchers to obtain, contribute, share, store, and analyze scientific algorithm through a portal platform. Due to incomplete existing scientific computing library, not so centralized massive knowledge resources, and distributed data analyses for a science team, it is desired that the development of SCLP as a centralized database for computing discipline could overcome the problems highlighted previously. Four disciplines of scientific computing are included in SCLP which are mathematics, engineering, computer science, and bioinformatics. Data collection was gathered from the expert in each discipline. This paper also highlights the importance of metadata to ease the process of obtaining the data. Result from this paper is about the architecture of scientific computing library portal for sequential algorithm which covers the four mentioned disciplines.

Web classification using extraction and machine learning techniques

Internet services that has become easier to access has contributed to the drastic increase in the number of web pages. This phenomenon has created new difficulties to internet users about retrieving the latest, relevant and excellent web information. This is due to the enormous contents of web information that have caused problems in the restructuring of web information. Thus, in order to ensure the latest, quality and relevant web information is optimally retrievable, it is necessary to undertake the task of web document classification. This paper discusses the result of classifying web document using the extraction and machine learning techniques. Four types of kernels namely the Radial Basis Function (RBF), linear, polynomial and sigmoid are applied to test the accuracy of the classification. The results show that the accuracy percentage of web document classification will increase whenever more web document is used. The results also show that linear kernel technique is the best in web document classification compared to RBF, polynomial and sigmoid.

End-User Perspectives on Effectiveness of Learning Performance Through Massive Open Online Course (MOOCs)

The phenomenon of using Massive Open Online Courses (MOOCs) in Malaysia higher education has raised questions regarding its effectiveness in improving the learning performance of students. The researchers have attempted to bridge the gap between MOOCs and learning performance in this study. Thus, this study has explored the factors that able to contribute to the use of MOOCs among university students by promoting learning performance. A total of 513 students received the survey and used in this study. Using the research model, seven predictors have proposed to study the effectiveness of MOOCs’ use. Upon analysis, the study revealed that students’ satisfaction, perceived enjoyment, and their attitude towards use are the strongest predictors in the context of MOOCs. Therefore, this paper provides insights into students’ use of MOOCs that leads to its effectiveness in improving their learning performance.

An Implementation of Metaheuristic Algorithms in Business Intelligence Focusing on Higher Education Case Study

The education sector has witnessed increasing interest in data-driven decision-making. Education sector requires the use of business intelligence (BI) to ensure the extraction of information allows the educational staff to function more effective. This paper illustrated the use of metaheuristic algorithms such as Genetic Algorithm (GA), Particle Swarm Optimization (PSO) and Ant Colony Optimization (ACO) in BI to ensure the selection of informative features for decision making. Higher education based case studies are discussed to prove that the proposed technique able to improve the decisions and results to select features that able to increase the number of postgraduates in graduating within time allocated. The research aimed to propose a novel method to identify and select informative features. The accuracy for proposed algorithm is ACO in this research is 96.2% while for GA is 83.1% and PSO is 93.3%. Experiments show that using the informative features has better analysis of data.

Analysis and Identification the Complexity of Data Heterogeneity on Learning Environment Using Ontology

Distributed and various systems on learning environment are the current issues to produce big data and heterogeneity data problem. Heterogeneity on learning environment is about numerous learning applications and various learning information to support a learning process in educational institutions. There are a lot of relationships are formed between elements on learning environment. The elements on learning environment consist of learning data, learning applications, data sources, learning concept, and data heterogeneity aspect on learning environment. These elements are interrelated and produce complex relationship between each other. A complex relationship problem between elements on learning environment makes a process of analysis and identification difficult to be done. Existing method to drawing this heterogeneity problem make confuse and misunderstanding readers. To solved this problem, researcher using ontology knowledge to describe and draw a semantic relationship that represent the complexity of data relationship on learning environment. The result of this analysis is to develop ontology knowledge to solve heterogeneity data problem specific in complexity relationship on learning environment. This result can give better understanding to the readers about complex relationship between elements on learning environment.

Parallel numerical solution for temperature prediction on laser glass cutting using AGE variances method

This paper discusses on the mathematical formulation in simulating the temperature behavior on a laser glass cutting. The mathematical simulation is solved by using three numerical methods, Alternating Group Explicit Method with Brian and Douglas variant, and Gauss Seidel Red Black method. This simulation is then being run on a parallel environment using the master and workers methodology where 20 units of processors work simultaneously in solving and visualizing the data. Lastly, the results obtained from the simulation is compared by using the parallel performance measurement and also on a large sparse matrix size.

Characteristics determination on Scientific Computing Library Portal

This paper presents the development of Scientific Computing Library Portal (SCLP) for scientific community application at Universiti Teknologi Malaysia (UTM). The purpose of developing SCLP is to overcome the arise problems in scientific domains. Three roles of users have been identified which are Common User, Scientific Algorithm Provider and Scientific Librarian. Several characteristics of scientific computing library have been well described. The results from the characteristics analysis are used to develop SCLP, as an improvement of the existing scientific computing libraries.