Magister Scientiae - MSc (Computer Science)

Permanent URI for this collection

https://hdl.handle.net/10566/14793

Browse

Now showing 1 - 20 of 94

Semantic data access for relational databases using an ontology
(University of the Western Cape, 2024) Jafta, Yahlieel; Leenen, Louise
Data analysis-based decision-making is performed daily by domain experts. As data grows in size and heterogeneity, accessing relevant data becomes challenging. In an Ontology-based data access (OBDA) approach, ontologies are advocated as a suitable formal tool to address complex data access. This technique falls within the Semantic Web domain, combining a domain ontology with a data source by using a declarative mapping specification to enable data access using a domain vocabulary. In this research, we investigate this approach by: a) studying the theoretical background that enables this technique; b) conducting a literature review on the existing open source tools that implement OBDA; c) implementing OBDA on a “real-world” relational dataset using an OBDA tool; and d) providing results and analysis of query answering. We selected Ontop (https://ontop-vkg.org) among various OBDA tools to illustrate how this technique enhances the data usage of the GitHub community. Ontop is an open-source tool applying OBDA in the domain of relational databases. We used the GHTorrent dataset, a relational database, in combination with the SemanGit ontology for our implementation.
On the efficacy of enhanced feature selection methods for supervised crime prediction
(University of the Western Cape, 2023) May, Sphamandla Innocent; Isafiade, Omowunmi
The challenge of crime across the globe has necessitated several considerations for crime preventive measures. There exist a variety of crime prevention strategies, such as the use of necessary weapons or tools to respond to crime. However, for resource-constrained nations such as South Africa, where the current police to civilian ratio is overwhelming, this may not suffice. Consequently, crime continues to be on the rise, necessitating alternative prevention strategies. Among alternative prevention approaches, the use of historical crime data can be explored through machine learning. Crime prediction using machine learning has been explored and has shown promising results. However, the choice of algorithm and feature selection methods play a critical role in creating an effective predictive model. This study, therefore, explores the efficacy of enhanced feature selection methods in supervised machine learning algorithms for crime prediction. Four (4) baseline algorithms are adopted, which are Random Forest (RF), Extremely Randomized Trees (ERT), Na¨ıve Bayes (NB), and Support Vector Machine (SVM). This research further proposes three algorithms, with the first derived from hybridizing RF and ERT (RF-Plus), while the other two (2) were obtained from enhancing NB and SVM using recursive feature elimination (RFE), obtaining (RFE-NB) and (RFE-SVM) respectively, totaling seven algorithms. Finally, a comparative evaluation of these algorithms with their respective baselines is conducted to report on their efficacy and contrasted against additional two (2) algorithms from the literature, which amounts to a total of nine (9) algorithms. The study conducted performance evaluation on the models using two distinct publicly available datasets, which are the Chicago and Los Angeles crime datasets. Results confirm that feature selection positively impacts prediction accuracy. The enhancement on the pure NB improved its accuracy from 72.5% to 96.6% and 80.45% to 95.78% for Chicago and Los Angeles datasets, respectively. The enhancement improved the accuracy of pure SVM from 74.73% to 89.91% and 75.73% to 88.70% for the Chicago and Los Angeles datasets, respectively, while achieving 97.04% and 95.5% on RF-Plus for both Chicago and Los Angeles datasets, respectively.
Exploring low-cost solution for 3D crime scene data gathering with immersive technology
(University of the Western Cape, 2023) Mfundo Andrew, Maneli; Isafiade, Omowunmi Elizabeth
3D crime scene data gathering is critical for law enforcement and investigators during crime scene investigations. Crime scene investigations have seen the effective usage of Light Detection and Ranging (LiDAR) scanners for 3D reconstruction alongside immersive technologies, such as Augmented Reality (AR) and Virtual Reality (VR). However, the inability to afford the existing high-end devices that can offer the desired accuracy of 3D scene data collection in low-resource settings cannot be overlooked, as this may impede crime investigations or render some crime cases insoluble.
Application of Several Time Series Methods to Three Important Financial Time Series
(University of the Western Cape, 2007) O'Connell, Bryan; Koean, C
This study is concerned with three different financial time series over an eight year period, namely: the government repurchase rate, the Rand-Dollar exchange rate and the Allshare Index. The aim is to better understand the statistical nature of the time series. The theory employed will be discussed briefly and then the results will be reported. Different methods are employed to model the different time series. The following topics are discussed: unit root tests, autoregressive integrated moving average models, outlier tests, transformations, generalised autoregressive conditional heteroscedasticity models, cointegration, transfer function models and vector autoregressive models.
Semi-synchronous video for deaf telephony with an adapted synchronous codec
(University of the Western Cape, 2009) Ma, Zhenyu; Tucker, William D.
As Information and Communication Technology (ICT) matures, communication services must be improved to meet the needs of all types of users. For some uses, current Video over Internet Protocol (IP) brings unsatisfactory and even unrecognisable quality of video sequences. Such communication does not always meet the needs of Deaf 1 people. Asynchronous video messaging, such as EyeJot (www.eyejot.com), offers Deaf people the ability to send and receive video messages like email. Unfortunately, communicating like this incurs much delay, resulting in slow response. Even though text messaging is popular among Deaf people via cellphone or Internet, but they would prefer to use sign language for communication. Video Relay Service (VRS) attempts to help Deaf users communicate with hearing people in sign language. VRS provides synchronous video and voice services to enable those who use sign language to communicate with hearing people through a relay interpreter across the world via the Internet.
Long short-term memory recurrent neural networks for signature verification
(UWC, 2003) Tiflin, C; Omlin, C
Handwritten signature verification is defined as the classification process that strives to learn the manner in which an individual makes use of the muscular memory of their hands, fingers, and wrist to reproduce a signature. A handwritten signature is captured by a pen input device and sampled at a high frequency which results in time series with several hundred data points. A novel recurrent neural network architecture known as long short-term memory was designed for modeling such a long-time series. This research investigates the suitability of long short-term memory recurrent neural networks for the task of online signature verification. We design and experiment with various network architectures to determine if this model can be trained to discriminate between authentic and fraudulent signatures. We further determine whether the complexity of a signature impacts the performance level of the network when applied to fraudulent signatures. We also investigate the performance level of the network when varying the number of signature features.
Towards a chereme based dynamic South African sign language gesture recognition system
(University of the Western Cape, 2007) Machanja, Addmore; Bajic, Vladimir B.
Hand gestures are a natural and intuitive way of human to human communication. Motivated by the achievements made towards automatic speech recognition, and by the ease with which people sign, many researchers started working on sign language recognition systems. Besides, technologies used to build gesture recognition systems pose as an alternative to the cumbersome and the failure prone mechanical devices that are currently used as human-machine interface devices. Most of the available gesture recognition systems represent each sign language gesture with an individual gesture model. Such systems can only recognize a limited number of dynamic sign language gestures. It is cumbersome to build and maintain a gesture recognition system that uses thousands and thousands of individual gesture models. Sign language linguists argue that all sign language gestures are derived from small sets of reusable components, the cheremes.
Automatic real-time facial expression recognition for signed language translation
(University of the Western Cape, 2006) Whitehill, Jacob Richard; Omlin, Christian W
We investigated two computer vision techniques designed to increase both the recognition accuracy and computational efficiency of automatic facial expression recognition. In particular, we compared a local segmentation of the face around the mouth, eyes, and brows to a global segmentation of the whole face. Our results indicated that, surprisingly, classifying features from the whole face yields greater accuracy despite the additional noise that the global data may contain. We attribute this in part to correlation effects within the Cohn-Kanade database. We also developed a system for detecting FACS action units based on Haar features and the Adaboost boosting algorithm. This method achieves equally high recognition accuracy for certain AUs but operates two orders of magnitude more quickly than the Gabor+SVM approach. Finally, we developed a software prototype of a real-time, automatic signed language recognition system using FACS as an intermediary framework.
Handwritten alphabet character recognition using audio signatures and machine learning
(University of the Western Cape, 2023) Beck, Bruce; Ghaziasgar, Mehrdad
This research investigates the creation of an audio-based character recognition system that is able to segment, process and recognise uppercase English letters continuously drawn by the user on a given writing surface such as a table-top using a generic writing implement. The aim is to make use of the microphones on a single smartphone to capture the acoustic signal generated by the user as they draw letters on the writing surface, followed by the application of audio segmentation to subdivide the audio signal into segments corresponding to each letter, and finally the application of a combination of the Mel-Frequency Cepstral Coefficients feature descriptor and Support Vector Machines to recognise the segmented letters.
A comparative evaluation of population-based optimization algorithms for workflow scheduling in cloud-fog environments
(University of the Western Cape, 2022) Subramoney, Dineshan; Nyirenda, Clement
Scientific workflows are denoted by interdependent tasks and computations that are aimed at achieving some scientific objectives. The scheduling of these workflows involve the allocation of the tasks to particular computational resources, traditionally on the cloud infrastructure. This process is, however, very challenging. It is associated with high computation and communication costs because scientific workflows are data-intensive and computationally complex. In recent years, there has been overwhelming interest in using population-based optimization algorithms such as Particle Swarm Optimization (PSO) and Genetic Algorithms (GA) for scientific workflow scheduling, predominantly, in the cloud environments.
Credit Card Transactions Fraud Detection, and Machine Learning: Modelling Time with LSTM Recurrent Neural Networks
(University of the Western Cape, 2007) Wiese, Benard Jacobus; Omlin, Christian W.
In recent years, topics such as fraud detection and fraud prevention have received a lot of attention on the research front, in particular from plastic card issuers. The reason for this increase in research activity can be attributed to the huge annual financial losses incurred by card issuers due to fraudulent use of their card products. A successful strategy for dealing with fraud can quite literally mean millions of dollars in savings per year on operational costs. Artificial neural networks have come to the front as an at least partially successful method for fraud detection. The success of neural networks in this field is, however, limited by their underlying design - a feedforward neural network is simply a static mapping of input vectors to output vectors, and as such is incapable of adapting to changing shopping profiles of legitimate card holders. Thus, fraud detection systems in use today are plagued by misclassifications and their usefulness is hampered by high false positive rates. We address this problem by proposing the use of a dynamic machine learning method in an attempt to model the time series inherent in sequences of same card transactions. We believe that, instead of looking at individual transactions; it makes more sense to look at sequences of transactions as a whole; a technique that can model time in this context will be more robust to minor shifts in legitimate shopping behaviour. In order to form a clear basis for comparison, we did some investigative research on feature selection, pre-processing, and on the selection of performance measures; the latter will facilitate comparison of results obtained by applying machine learning methods to the biased data sets largely associated with fraud detection. We ran experiments on real world credit card transactional data using three machine learning techniques: a conventional feedforward neural network (FFNN), and two innovative methods, the support vector machine (SVM) and the long short-term memory recurrent neural network (LSTM).
Geo-physical parameter forecasting on imagery{based data sets using machine learning techniques
(University of Western Cape, 2021) Hussein, Eslam; Ghaziasgar, Mehrdad
This research objectively investigates the e ectiveness of machine learning (ML) tools towards predicting several geo-physical parameters. This is based on a large number of studies that have reported high levels of prediction success using ML in the eld. Therefore, several widely used ML tools coupled with a number of di erent feature sets are used to predict six geophysical parameters namely rainfall, groundwater, evapora- tion, humidity, temperature, and wind. The results of the research indicate that: a) a large number of related studies in the eld are prone to speci c pitfalls that lead to over-estimated results in favour of ML tools; b) the use of gaussian mixture models as global features can provide a higher accuracy compared to other local feature sets; c) ML never outperform simple statistically-based estimators on highly-seasonal parame- ters, and providing error bars is key to objectively evaluating the relative performance of the ML tools used; and d) ML tools can be e ective for parameters that are slow- changing such as groundwater.
KernTune: Self-tuning Linux Kernel Performance Using Support Vector Machines
(University of the Western Cape, 2006) Yi, Long; Connan, James
Self-tuning has been an elusive goal for operating systems and is becoming a pressing issue for modern operating systems. Well-trained system administrators are able to tune an operating system to achieve better system performance for a specific system class. Unfortunately, the system class can change when the running applications change. Our model for self-tuning operating system is based on a monitor-classify- adjust loop. The idea of this loop is to continuously monitor certain performance metrics, and whenever these change, the system determines the new system class and dynamically adjusts tuning parameters for this new class. This thesis describes KernTune, a prototype tool that identifies the system class and improves system performance automatically. A key aspect of KernTune is the notion of Artificial Intelligence (AI) oriented performance tuning. It uses a support vector machine (SVM) to identify the system class, and tunes the operating system for that specific system class. This thesis presents design and implementation details for KernTune. It shows how KernTune identifies a system class and tunes the operating system for improved performance.
The Efficacy of the Eigenvector Approach to South African Sign Language Identification
(University of the Western Cape, 2010) Segers, Vaughn Mackman; Connan, James
The communication barriers between deaf and hearing society mean that interaction between these communities is kept to a minimum. The South African Sign Language research group, Integration of Signed and Verbal Communication: South African Sign Language Recognition and Animation (SASL), at the University of the Western Cape aims to create technologies to bridge the communication gap. In this thesis we address the subject of whole hand gesture recognition. We demonstrate a method to identify South African Sign Language classifiers using an eigenvector approach. The classifiers researched within this thesis are based on those outlined by the Thibologa Sign Language Institute for SASL. Gesture recognition is achieved in real time. Utilising a pre-processing method for image registration we are able to increase the recognition rates for the eigenvector approach.
A comparative evaluation of 3d and spatio-temporal deep learning techniques for crime classification and prediction
(University of Western Cape, 2021) Matereke, Tawanda Lloyd; Ghaziasgar, Mehrdad
This research is on a comparative evaluation of 3D and spatio-temporal deep learning methods for crime classification and prediction using the Chicago crime dataset, which has 7.29 million records, collected from 2001 to 2020. In this study, crime classification experiments are carried out using two 3D deep learning algorithms, i.e., 3D Convolutional Neural Network and the 3D Residual Network. The crime classification models are evaluated using accuracy, F1 score, Area Under Receiver Operator Curve (AUROC), and Area Under Curve - Precision-Recall (AUCPR). The effectiveness of spatial grid resolutions on the performance of the classification models is also evaluated during training, validation and testing.
Chereme- Based Recognition of Isolated, Dynamic Gestures from South African Sign Language with Hidden Markov Models
(University of the Western Cape, 2006) Rajah, Christopher; Omlin, Christian
Much work has been done in building systems that can recognise gestures, e.g. as a component of sign language recognition systems. These systems typically use whole gestures as the smallest unit for recognition. Although high recognition rates have been reported, these systems do not scale well and are computationally intensive. The reason why these systems generally scale poorly is that they recognize gestures by building individual models for each separate gesture; as the number of gestures grows, so does the required number of models. Beyond a certain threshold number of gestures to be recognized, this approach becomes infeasible. This work proposes that similarly good recognition rates can be achieved by building models for subcomponents of whole gestures, so-called cheremes. Instead of building models for entire gestures, we build models for cheremes and recognize gestures as sequences of such cheremes. The assumption is that many gestures share cheremes and that the number of cheremes necessary to describe gestures is much smaller than the number of gestures. This small number of cheremes then makes it possible to recognize a large number of gestures with a small number of chereme models. This approach is akin to phoneme-based speech recognition systems where utterances are recognized as phonemes which in turn are combined into words. We attempt to recognise and classify cheremes found in South African Sign Language (SASL). We introduce a method for the automatic discovery of cheremes in dynamic signs. We design, train and use hidden Markov models (HMMs) for chereme recognition. Our results show that this approach is feasible in that it not only scales well, but it also generalizes well. We are able to recognize cheremes in signs that were not used for training HMMs; this generalization ability is a basic necessity for chemere-based gesture recognition. Our approach can thus lay the foundation for building a SASL dynamic gesture recognition system.
A Digital Identity Management System
(University of the Western Cape, 2007) Phiri, Jackson; Agbinya, Johnson
The recent years have seen an increase in the number of users accessing online services using communication devices such as computers, mobile phones and cards based credentials such as credit cards. This has prompted most governments and business organizations to change the way they do business and manage their identity information. The coming of the online services has however made most Internet users vulnerable to identity fraud and theft. This has resulted in a subsequent increase in the number of reported cases of identity theft and fraud, which is on the increase and costing the global industry excessive amounts. Today with more powerful and effective technologies such as artificial intelligence, wireless communication, mobile storage devices and biometrics, it should be possible to come up with a more effective multi-modal authentication system to help reduce the cases of identity fraud and theft. A multi-modal digital identity management system IS proposed as a solution for managing digital identity information in an effort to reduce the cases of identity fraud and theft seen on most online services today. The proposed system thus uses technologies such as artificial intelligence and biometrics on the current unsecured networks to maintain the security and privacy of users and service providers in a transparent, reliable and efficient way. In order to be authenticated in the proposed multi-modal authentication system, a user is required to submit more than one credential attribute. An artificial intelligent technology is used to implement a technique of information fusion to combine the user's credential attributes for optimum recognition. The information fusion engine is then used to implement the required multi-modal authentication system.
South African Sign Language Recognition Using Feature Vectors and Hidden Markov Models
(University of the Western Cape, 2010) Naidoo, Nathan Lyle; Connan, James
This thesis presents a system for performing whole gesture recognition for South African Sign Language. The system uses feature vectors combined with Hidden Markov models. In order to construct a feature vector, dynamic segmentation must occur to extract the signer's hand movements. Techniques and methods for normalising variations that occur when recording a signer performing a gesture, are investigated. The system has a classification rate of 69%.
Security related self-protected networks: Autonomous threat detection and response (ATDR)
(University of Western Cape, 2021) Havenga, Wessel Johannes Jacobus; Bagula, Bigomokero
Cybersecurity defense tools, techniques and methodologies are constantly faced with increasing challenges including the evolution of highly intelligent and powerful new-generation threats. The main challenges posed by these modern digital multi-vector attacks is their ability to adapt with machine learning. Research shows that many existing defense systems fail to provide adequate protection against these latest threats. Hence, there is an ever-growing need for self-learning technologies that can autonomously adjust according to the behaviour and patterns of the offensive actors and systems. The accuracy and effectiveness of existing methods are dependent on decision making and manual input by human experts. This dependence causes 1) administration overhead, 2) variable and potentially limited accuracy and 3) delayed response time.
Self-tuning Linux Kernel Performance Using Support Vector Machines
(University of the Western Cape, 2006) Yi, Long; Connan, James
In this chapter, we provide the motivation and background behind the automatic optimisation of an operating system. We begin with a discussion of some of the difficulties of automatic operating system optimisation and the benefits of automatic optimisation technology which inspired our research. We then describe the research problem and aims. Thereafter, our approach and methodology are explained. Finally, the organisation of the thesis and summary are presented. 1.1 Background and Motivation In today's networking world, a mission-critical server requires consistently good performance [2] . To this end, almost all operating systems which run on such a critical server are managed by system administrators who should be skillful and experienced in tuning operating systems by adjusting system configuration and performance parameters of the operating system to run a specific system workload. This involves system capacity planning, performance metrics, workload characteristics, system settings, etc. Skillful system administrators are scarce and expensive. As computer hardware becomes cheaper and free critical computer software becomes more viable, e.g., Linux, Samba, Mysql, Apache, the total cost of ownership for building and maintaining a mission-critical server becomes more and more dominated by the cost of human resources. Furthermore, with the increasing number of new applications and services, a modern operating system offers more system parameters with larger ranges for more system classes than ever before. This situation serves as our motivation for a new generation of automatic optimisation technology for operating systems. The potential benefits of the automatic optimisation technology will be amplified as future applications and operating systems become more complex.

Browse

collection.page.browse.recent.head