arXiv subjects

Tshilidzi Marwala

Publications and source records attributed to Tshilidzi Marwala.

At least 55 records · Page 3Linked to original sources

arXiv2011

Improving the performance of the ripper in insurance risk classification : A comparitive study using feature selection

The Ripper algorithm is designed to generate rule sets for large datasets with many features. However, it was shown that the algorithm struggles with classification performance in the presence of missing data. The algorithm struggles to classify instances when the quality of the data deteriorates as a result of increasing missing data. In this paper, a feature selection technique is used to help improve the classification performance of the Ripper model. Principal component analysis and evidence automatic relevance determination techniques are used to improve the performance. A comparison is done to see which technique helps the algorithm improve the most. Training datasets with completely observable data were used to construct the model and testing datasets with missing values were used for measuring accuracy. The results showed that principal component analysis is a better feature selection for the Ripper in improving the classification performance.

cs.LG↗

arXiv2010

Application of Global and One-Dimensional Local Optimization to Operating System Scheduler Tuning

This paper describes a study of comparison of global and one-dimensional local optimization methods to operating system scheduler tuning. The operating system scheduler we use is the Linux 2.6.23 Completely Fair Scheduler (CFS) running in simulator (LinSched). We have ported the Hackbench scheduler benchmark to this simulator and use this as the workload. The global optimization approach we use is Particle Swarm Optimization (PSO). We make use of Response Surface Methodology (RSM) to specify optimal parameters for our PSO implementation. The one-dimensional local optimization approach we use is the Golden Section method. In order to use this approach, we convert the scheduler tuning problem from one involving setting of three parameters to one involving the manipulation of one parameter. Our results show that the global optimization approach yields better response but the one- dimensional optimization approach converges to a solution faster than the global optimization approach.

cs.OS↗

arXiv2010

Use of Data Mining in Scheduler Optimization

The operating system's role in a computer system is to manage the various resources. One of these resources is the Central Processing Unit. It is managed by a component of the operating system called the CPU scheduler. Schedulers are optimized for typical workloads expected to run on the platform. However, a single scheduler may not be appropriate for all workloads. That is, a scheduler may schedule a workload such that the completion time is minimized, but when another type of workload is run on the platform, scheduling and therefore completion time will not be optimal; a different scheduling algorithm, or a different set of parameters, may work better. Several approaches to solving this problem have been proposed. The objective of this survey is to summarize the approaches based on data mining, which are available in the literature. In addition to solutions that can be directly utilized for solving this problem, we are interested in data mining research in related areas that have potential for use in operating system scheduling. We also explain general technical issues involved in scheduling in modern computers, including parallel scheduling issues related to multi-core CPUs. We propose a taxonomy that classifies the scheduling approaches we discuss into different categories.

cs.OS↗

arXiv2009

State of the Art Review for Applying Computational Intelligence and Machine Learning Techniques to Portfolio Optimisation

Computational techniques have shown much promise in the field of Finance, owing to their ability to extract sense out of dauntingly complex systems. This paper reviews the most promising of these techniques, from traditional computational intelligence methods to their machine learning siblings, with particular view to their application in optimising the management of a portfolio of financial instruments. The current state of the art is assessed, and prospective further work is assessed and recommended

cs.CE↗

arXiv2009

Finite element model selection using Particle Swarm Optimization

This paper proposes the application of particle swarm optimization (PSO) to the problem of finite element model (FEM) selection. This problem arises when a choice of the best model for a system has to be made from set of competing models, each developed a priori from engineering judgment. PSO is a population-based stochastic search algorithm inspired by the behaviour of biological entities in nature when they are foraging for resources. Each potentially correct model is represented as a particle that exhibits both individualistic and group behaviour. Each particle moves within the model search space looking for the best solution by updating the parameters values that define it. The most important step in the particle swarm algorithm is the method of representing models which should take into account the number, location and variables of parameters to be updated. One example structural system is used to show the applicability of PSO in finding an optimal FEM. An optimal model is defined as the model that has the least number of updated parameters and has the smallest parameter variable variation from the mean material properties. Two different objective functions are used to compare performance of the PSO algorithm.

cs.AI↗

arXiv2009

Foundations for a Developmental State: A case for technical education

This paper studies the viability of making a country a developmental state. In particular it studies the characteristics of a developmental state and how they are linked to technology. It then identifies technical education, as a vital force for the creation of a developmental state. In particular it identifies analytical, numeracy, computational and communication skills as vital forces for a society to create a developmental society.

cs.CY↗

arXiv2008

Prediction of Platinum Prices Using Dynamically Weighted Mixture of Experts

Neural networks are powerful tools for classification and regression in static environments. This paper describes a technique for creating an ensemble of neural networks that adapts dynamically to changing conditions. The model separates the input space into four regions and each network is given a weight in each region based on its performance on samples from that region. The ensemble adapts dynamically by constantly adjusting these weights based on the current performance of the networks. The data set used is a collection of financial indicators with the goal of predicting the future platinum price. An ensemble with no weightings does not improve on the naive estimate of no weekly change; our weighting algorithm gives an average percentage error of 63% for twenty weeks of prediction.

cs.AI↗

arXiv2008

Evaluating the Impact of Missing Data Imputation through the use of the Random Forest Algorithm

This paper presents an impact assessment for the imputation of missing data. The data set used is HIV Seroprevalence data from an antenatal clinic study survey performed in 2001. Data imputation is performed through five methods: Random Forests, Autoassociative Neural Networks with Genetic Algorithms, Autoassociative Neuro-Fuzzy configurations, and two Random Forest and Neural Network based hybrids. Results indicate that Random Forests are superior in imputing missing data in terms both of accuracy and of computation time, with accuracy increases of up to 32% on average for certain variables when compared with autoassociative networks. While the hybrid systems have significant promise, they are hindered by their Neural Network components. The imputed data is used to test for impact in three ways: through statistical analysis, HIV status classification and through probability prediction with Logistic Regression. Results indicate that these methods are fairly immune to imputed data, and that the impact is not highly significant, with linear correlations of 96% between HIV probability prediction and a set of two imputed variables using the logistic regression analysis.

stat.ME↗

arXiv2008

Comparison of Data Imputation Techniques and their Impact

Missing and incomplete information in surveys or databases can be imputed using different statistical and soft-computing techniques. This paper comprehensively compares auto-associative neural networks (NN), neuro-fuzzy (NF) systems and the hybrid combinations the above methods with hot-deck imputation. The tests are conducted on an eight category antenatal survey and also under principal component analysis (PCA) conditions. The neural network outperforms the neuro-fuzzy system for all tests by an average of 5.8%, while the hybrid method is on average 15.9% more accurate yet 50% less computationally efficient than the NN or NF systems acting alone. The global impact assessment of the imputed data is performed by several statistical tests. It is found that although the imputed accuracy is high, the global effect of the imputed data causes the PCA inter-relationships between the dataset to become altered. The standard deviation of the imputed dataset is on average 36.7% lower than the actual dataset which may cause an incorrect interpretation of the results.

stat.ME↗

arXiv2008

A note on the separability index

In discriminating between objects from different classes, the more separable these classes are the less computationally expensive and complex a classifier can be used. One thus seeks a measure that can quickly capture this separability concept between classes whilst having an intuitive interpretation on what it is quantifying. A previously proposed separability measure, the separability index (SI) has been shown to intuitively capture the class separability property very well. This short note highlights the limitations of this measure and proposes a slight variation to it by combining it with another form of separability measure that captures a quantity not covered by the Separability Index.

stat.ME↗

arXiv2008

Artificial Intelligence Techniques for Steam Generator Modelling

This paper investigates the use of different Artificial Intelligence methods to predict the values of several continuous variables from a Steam Generator. The objective was to determine how the different artificial intelligence methods performed in making predictions on the given dataset. The artificial intelligence methods evaluated were Neural Networks, Support Vector Machines, and Adaptive Neuro-Fuzzy Inference Systems. The types of neural networks investigated were Multi-Layer Perceptions, and Radial Basis Function. Bayesian and committee techniques were applied to these neural networks. Each of the AI methods considered was simulated in Matlab. The results of the simulations showed that all the AI methods were capable of predicting the Steam Generator data reasonably accurately. However, the Adaptive Neuro-Fuzzy Inference system out performed the other methods in terms of accuracy and ease of implementation, while still achieving a fast execution time as well as a reasonable training time.

cs.AI↗

arXiv2008

Social Learning Methods in Board Games

This paper discusses the effects of social learning in training of game playing agents. The training of agents in a social context instead of a self-play environment is investigated. Agents that use the reinforcement learning algorithms are trained in social settings. This mimics the way in which players of board games such as scrabble and chess mentor each other in their clubs. A Round Robin tournament and a modified Swiss tournament setting are used for the training. The agents trained using social settings are compared to self play agents and results indicate that more robust agents emerge from the social training setting. Higher state space games can benefit from such settings as diverse set of agents will have multiple strategies that increase the chances of obtaining more experienced players at the end of training. The Social Learning trained agents exhibit better playing experience than self play agents. The modified Swiss playing style spawns a larger number of better playing agents as the population size increases.

cs.AI↗

arXiv2008

Emergency Centre Organization and Automated Triage System

The excessive rate of patients arriving at accident and emergency centres is a major problem facing South African hospitals. Patients are prioritized for medical care through a triage process. Manual systems allow for inconsistency and error. This paper proposes a novel system to automate accident and emergency centre triage and uses this triage score along with an artificial intelligence estimate of patient-doctor time to optimize the queue order. A fuzzy inference system is employed to triage patients and a similar system estimates the time but adapts continuously through fuzzy Q-learning. The optimal queue order is found using a novel procedure based on genetic algorithms. These components are integrated in a simple graphical user interface. Live tests could not be performed but simulations reveal that the average waiting time can be reduced by 48 minutes and priority is given to urgent patients

cs.CY↗

arXiv2008

Bayesian evidence for finite element model updating

This paper considers the problem of model selection within the context of finite element model updating. Given that a number of FEM updating models, with different updating parameters, can be designed, this paper proposes using the Bayesian evidence statistic to assess the probability of each updating model. This makes it possible then to evaluate the need for alternative updating parameters in the updating of the initial FE model. The model evidences are compared using the Bayes factor, which is the ratio of evidences. The Jeffrey scale is used to determine the differences in the models. The Bayesian evidence is calculated by integrating the likelihood of the data given the model and its parameters over the a priori model parameter space using the new nested sampling algorithm. The nested algorithm samples this likelihood distribution by using a hard likelihood-value constraint on the sampling region while providing the posterior samples of the updating model parameters as a by-product. This method is used to calculate the evidence of a number of plausible finite element models.

stat.CO↗

arXiv2008

An Intelligent Multi-Agent Recommender System for Human Capacity Building

This paper presents a Multi-Agent approach to the problem of recommending training courses to engineering professionals. The recommendation system is built as a proof of concept and limited to the electrical and mechanical engineering disciplines. Through user modelling and data collection from a survey, collaborative filtering recommendation is implemented using intelligent agents. The agents work together in recommending meaningful training courses and updating the course information. The system uses a users profile and keywords from courses to rank courses. A ranking accuracy for courses of 90% is achieved while flexibility is achieved using an agent that retrieves information autonomously using data mining techniques from websites. This manner of recommendation is scalable and adaptable. Further improvements can be made using clustering and recording user feedback.

cs.AI↗

arXiv2008

Stochastic Optimization Approaches for Solving Sudoku

In this paper the Sudoku problem is solved using stochastic search techniques and these are: Cultural Genetic Algorithm (CGA), Repulsive Particle Swarm Optimization (RPSO), Quantum Simulated Annealing (QSA) and the Hybrid method that combines Genetic Algorithm with Simulated Annealing (HGASA). The results obtained show that the CGA, QSA and HGASA are able to solve the Sudoku puzzle with CGA finding a solution in 28 seconds, while QSA finding a solution in 65 seconds and HGASA in 1.447 seconds. This is mainly because HGASA combines the parallel searching of GA with the flexibility of SA. The RPSO was found to be unable to solve the puzzle.

cs.NE↗

arXiv2008

The Effect of Structural Diversity of an Ensemble of Classifiers on Classification Accuracy

This paper aims to showcase the measure of structural diversity of an ensemble of 9 classifiers and then map a relationship between this structural diversity and accuracy. The structural diversity was induced by having different architectures or structures of the classifiers The Genetical Algorithms (GA) were used to derive the relationship between diversity and the classification accuracy by evolving the classifiers and then picking 9 classifiers out on an ensemble of 60 classifiers. It was found that as the ensemble became diverse the accuracy improved. However at a certain diversity measure the accuracy began to drop. The Kohavi-Wolpert variance method is used to measure the diversity of the ensemble. A method of voting is used to aggregate the results from each classifier. The lowest error was observed at a diversity measure of 0.16 with a mean square error of 0.274, when taking 0.2024 as maximum diversity measured. The parameters that were varied were: the number of hidden nodes, learning rate and the activation function.

cs.LG↗

arXiv2008

Introduction to Relational Networks for Classification

The use of computational intelligence techniques for classification has been used in numerous applications. This paper compares the use of a Multi Layer Perceptron Neural Network and a new Relational Network on classifying the HIV status of women at ante-natal clinics. The paper discusses the architecture of the relational network and its merits compared to a neural network and most other computational intelligence classifiers. Results gathered from the study indicate comparable classification accuracies as well as revealed relationships between data features in the classification data. Much higher classification accuracies are recommended for future research in the area of HIV classification as well as missing data estimation.

cs.LG↗