Shawkat Ali

A meta-learning approach to automatic kernel selection for support vector machines

Appropriate choice of a kernel is the most important ingredient of the kernel-based learning methods such as support vector machine (SVM). Automatic kernel selection is a key issue given the number of kernels available, and the current trial-and-error nature of selecting the best kernel for a given problem. This paper introduces a new method for automatic kernel selection, with empirical results based on classification. The empirical study has been conducted among five kernels with 112 different classification problems, using the popular kernel based statistical learning algorithm SVM. We evaluate the kernels’ performance in terms of accuracy measures. We then focus on answering the question: which kernel is best suited to which type of classification problem? Our meta-learning methodology involves measuring the problem characteristics using classical, distance and distribution-based statistical information. We then combine these measures with the empirical results to present a rule-based method to select the most appropriate kernel for a classification problem. The rules are generated by the decision tree algorithm C5.0 and are evaluated with 10 fold cross validation. All generated rules offer high accuracy ratings.

Improved support vector machine generalization using normalized input space

Data pre-processing always plays a key role in learning algorithm performance. In this research we consider data pre-processing by normalization for Support Vector Machines (SVMs). We examine the normalization affect across 112 classification problems with SVM using the rbf kernel. We observe a significant classification improvement due to normalization. Finally we suggest a rule based method to find when normalization is necessary for a specific classification problem. The best normalization method is also automatically selected by SVM itself.

Energy-efficient TDMA MAC protocol for wireless sensor networks applications

The availability of low-powered and cheap microprocessors, radio frequency integrated circuits and the development of new wireless communication techniques, make the wireless sensor networks (WSN) one of todays most promising technologies. Minimizing energy consumption and maximizing the lifetime of the networks are key requirements in the design of sensor network applications. Optimally designed medium access control (MAC) and routing protocols minimize energy consumption and prolong the network life. In this study, we have investigated an energy-efficient adaptive TDMA (EA-TDMA) protocol for railway applications that used in communication between sensor nodes and the cluster-head (CH) placed in a railway wagon. This protocol is suitable for medium traffic applications and reduces energy consumption by shortening the idle period when devices have no data to transmit. We have developed an analytical model for EA-TDMA and compared its performance with conventional TDMA and bit-map-assisted (BMA) protocols.

On optimal degree selection for polynomial kernel with support vector machines: Theoretical and empirical investigations

The key challenge in kernel based learning algorithms is the choice of an appropriate kernel and its optimal parameters. Selecting the optimal degree of a polynomial kernel is critical to ensure good generalisation of the resulting support vector machine model. In this paper we propose Bayesian and Laplace approximation methods to estimate the polynomial degree. A rule based meta-learning approach is then proposed for automatic polynomial kernel and its optimal degree selection. The new approach is constructed and tested on different sizes of 112 datasets with binary class as well as multi class classification problems. An extensive computational evaluation of these methods is conducted, and rules are generated to determine when these approximation methods are appropriate.

Reduction of power consumption in sensor network applications using machine learning techniques

Wireless sensor networking (WSN) and modern machine learning techniques have encouraged interest in the development of vehicle monitoring systems that ensure safe and secure operations of the rail vehicle. To make an energy-efficient WSN application, power consumption due to raw data collection and pre-processing needs to be kept to a minimum level. In this paper, an energy-efficient data acquisition method has investigated for WSN applications using modern machine learning techniques. In an existing system, four sensor nodes were placed in each railway wagon to collect data to develop a monitoring system for railways. In this system, three sensor nodes were placed in each wagon to collect the same data using popular regression algorithms, which reduces power consumption of the system. This study was conducted using six different regression algorithms with five different datasets. Finally the best suitable algorithm have suggested based on the performance metrics of the algorithms that include: correlation coefficient, root mean square error (RMSE), mean absolute error (MAE), root relative squared error (RRSE), relative absolute error (RAE) and computation complexity.

Comparative Performance of Nonlinear Distortion Effects in an OFDM-RoF Link

Recently, the integration of orthogonal frequency division multiplexing (OFDM) and radio over fiber (RoF) technique emerged the possibility of cost-effective and high data rate ubiquitous wireless networks. However, the nonlinear effects can severely degrade the performance of RoF as well as OFDM system. In this paper, the performance of OFDM-based RoF link with Mach-Zehnder modulator nonlinearity effects have been analyzed and simulated and then compared with single-carrier modulation case. Later, offset biased pre-distortion technique is applied in the OFDM-RoF link to produce a pre-distorted Mach-Zehnder modulator input voltage. Simulation results showed important observations in terms of distortion-free dynamic range and signal to noise ratio (SNR) for the applied offset biased pre-distortion technique

Retrospective Analysis for Mining the Causes in Manufacturing Processes

There has been a considerable growth in the use of statistical process control (SPC) for improving the quality in business, industries, or software development since the last decade. However, the processes are growing much more complex, and there is a tremendous increase of data size owning to the use of automated record machine. The conventional SPC tools become less effective in analyzing and identifying the cause of the process failures. This paper extends the idea of the modified centered CUSUMS, and proposes a new data selection procedure so that the associative discovery technique can be used in retrospective SPC analysis. Through our approach, the common data mining method (i.e. associative discovery) can be used to find the hidden knowledge from the data, and identify the causes of the process failure or success for the quality improvement. Besides, the hidden information that we extracted from the data can be used as supplement for the cause and effect diagram in the on-line process control.