Abdullah Embong

CLUSTERING HIGH DIMENSIONAL DATA USING SUBSPACE AND PROJECTED CLUSTERING ALGORITHMS

Problem statement: Clustering has a number of techniques that have been developed in statistics, pattern recognition, data mining, and other fields. Subspace clustering enumerates clusters of objects in all subspaces of a dataset. It tends to produce many over lapping clusters. Approach: Subspace clustering and projected clustering are research areas for clustering in high dimensional spaces. In this research we experiment three clustering oriented algorithms, PROCLUS, P3C and STATPC. Results: In general, PROCLUS performs better in terms of time of calculation and produced the least number of un-clustered data while STATPC outperforms PROCLUS and P3C in the accuracy of both cluster points and relevant attributes found. Conclusions/Recommendations: In this study, we analyze in detail the properties of different data clustering method.

Feature selection and classification of breast cancer diagnosis based on support vector machines

Support Vector Machines (SVM) is a new algorithm of data mining technique, recently received increasing popularity in machine learning community. This paper emphasizes how 1-norm SVM can be used in feature selection and smooth SVM (SSVM) for classification. As a case study, a breast cancer diagnosis was implemented. First, feature selection for support vector machines was utilized to determine the important features. Then, SSVM was used to classify the state of disease (benign or malignant) of breast cancer. As a result, SVM can achieve the state of the art performance on feature selection and classification.

Applying Kernel Logistic Regression in data mining to classify credit risk

Credit risk evaluation is an interesting and important data mining problem in financial analysis domain. This problem domain, do require estimable class probabilities as well as accurate classification method. One of classification methods in the kernel-machine techniques and data mining communities that allows non linear probabilistic classification, transparent reasoning, and competitive discriminative ability is Kernel Logistic Regression. Kernel Logistic Regression model is a kernelized version of Logistic Regression, which well known classification method in the field of statistical learning. The parameters of kernel model are given by the solution of a convex optimization problem, that can be found using the efficient Iteratively Re-weighted Least Squares (IRLS) algorithm. In this paper, we investigated the classification performance of applying Kernel Logistic Regression to classify risk credit problem. The result demonstrated that Kernel Logistic Regression has good accuracy to evaluate credit risk, comparable with another well known kernel machine, Support Vector Machine.

Data Mining Technique for Medical Diagnosis Using a New Smooth Support Vector Machine

In last decade, the uses of data mining techniques in medical studies are growing gradually. The aim of this paper is to present a recent research on the application of data mining technique for medical diagnosis problems. The proposed data mining technique is Multiple Knot Spline Smooth Support Vector Machine (MKS-SSVM). MKS-SSVM is a new SSVM which used multiple knot spline function to approximate the plus function instead the integral sigmoid function in SSVM. To evaluate the effectiveness of our method, we carried out on two medical dataset (diabetes disease and heart disease). The accuracy of previous results of these data still under 90% so far. The results of this study showed that MKS-SSVM was effective to diagnose medical dataset, especially diabetes disease and heart disease and this is very promising result compared to the previously reported results.

A new expert system for diabetes disease diagnosis using modified spline smooth support vector machine

In recent years, the uses of intelligent methods in biomedical studies are growing gradually. In this paper, a novel method for diabetes disease diagnosis using modified spline smooth support vector machine (MS-SSVM) is presented. To obtain optimal accuracy results, we used Uniform Design method for selection parameter. The performance of the method is evaluated using 10-fold cross validation accuracy, confusion matrix, sensitivity and specificity. The comparison with previous spline SSVM in diabetes disease diagnosis also was given. The obtained classification accuracy using 10-fold cross validation is 96.58%. The results of this study showed that the modified spline SSVM was effective to detect diabetes disease diagnosis and this is very promising result compared to the previously reported results.

Alternative Model for Extracting Multidimensional Data Based-On Comparative Dimension Reduction

In line with the technological developments, the current data tends to be multidimensional and high dimensional, which is more complex than conventional data and need dimension reduction. Dimension reduction is important in cluster analysis and creates a new representation for the data that is smaller in volume and has the same analytical results as the original representation. To obtain an efficient processing time while clustering and mitigate curse of dimensionality, a clustering process needs data reduction. This paper proposes an alternative model for extracting multidimensional data clustering based on comparative dimension reduction. We implemented five dimension reduction techniques such as ISOMAP (Isometric Feature Mapping), KernelPCA, LLE (Local Linear Embedded), Maximum Variance Unfolded (MVU), and Principal Component Analysis (PCA). The results show that dimension reductions significantly shorten processing time and increased performance of cluster. DBSCAN within Kernel PCA and Super Vector within Kernel PCA have highest cluster performance compared with cluster without dimension reduction.

A Framework of Dashboard System for Higher Education Using Graph-Based Visualization Technique

We propose a novel approach of knowledge visualization method by adopting graph-based visualization technique and incorporating Dashboard concept for higher education institutions. Two aspects are emphasized, knowledge visualization and human-machine interaction. The knowledge visualization helps users to analyze the comprehensive characteristics of the students, lecturers and subjects after the clustering process and the interaction enable domain knowledge transfer and the use of the human’s perceptual capabilities, thus increases the intelligence of the system. The knowledge visualization is enhanced through the dashboard concept where it provides significant patterns of knowledge on real-world and theoretical modeling which could be called wisdom. The framework consists of the dashboard model, system architecture and system prototype for higher education environment is presented in this paper.

Reduced Support Vector Machine Based on k-Mode Clustering for Classification Large Categorical Dataset

The smooth support vector machine (SSVM) is one of the promising algorithms for classification problems. However, it is restricted to work well on a small to moderate dataset. There exist computational difficulties when we use SSVM with non linear kernel to deal with large dataset. Based on SSVM, the reduced support vector machine (RSVM) was proposed to solve these difficulties using a randomly selected subset of data to obtain a nonlinear separating surface. In this paper, we propose an alternative algorithm, k-mode RSVM (KMO-RSVM) that combines RSVM with k-mode clustering technique to handle classification problems on categorical large dataset. In our experiments, we tested the effectiveness of KMO-RSVM on four public available dataset. It turns out that KMO-RSVM can improve speed of running time significantly than SSVM and still obtained a high accuracy. Comparison with RSVM indicates that KMO-RSVM is faster, gets smaller reduced set and comparable testing accuracy than RSVM.

Computation offloading for smart internet devices

With the incessant advancement of smart internet devices and their ubiquitous usage, users prospect the same performance as if they used to run application on resources rich desktop devices. Smart internet devices are poor in resources such as storage, capacity, processing performance and battery life. Such mobile devices deliver lower performance as they are constrained by weight, size and mobility despite all the developments. These limitations can be ameliorated by utilizing the technique known as computation offloading. Computation offloading is the process of transferring compute-intensive data to resources rich servers called surrogates to run the entire or parts of application on behalf of mobile devices. In this paper, we present a survey on steps, criterion, types, flow and necessity of computation offloading with the reviewed computation offloading schemes. The Paper also proposes suggestions and opinions for future work in related field.

Credit risk classification using Kernel Logistic Regression with optimal parameter

Recently, Machine Learning techniques have become very popular because of its effectiveness. This study, applies Kernel Logistic Regression (KLR) to the credit risk classification in an attempt to suggest a model with better classification accuracy. Credit risk classification is an interesting and important data mining problem in financial analysis domain. In this study, the optimal parameter values (regularization and kernel function) of KLR. are found by using a grid search technique with 5-fold cross-validation. Credit risk data sets from UCI machine learning are used in order to verify the effectiveness of the KLR method in classifying credit risk. The experiment results show that KLR has promising performance when compared with other Machine Learning techniques in previous research literatures.