SAICSIT 2022: Papers with Abstracts

Abstract. In recent years cybersecurity challenges and concerns have become a common theme for discussion by both the government and private sector. These challenges are partly brought on by the continued use of and dependence on information technology, such as the internet, wireless networks and the development and use of smart devices. Additionally, the Covid-19 pandemic has also led to the increase in internet use as it altered the way in which people live and work through forcing businesses and even schools to move to remote working. All these events have made cybersecurity challenges and concerns spiral and more so in Africa where cybercrime continues to rise and be a constant threat. This study proposes a cybersecurity community of practice as a strategy to address African contextual cybersecurity challenges. This qualitative enquiry, based on organizations on the African continent, identifies key characteristics and objectives of an African cybersecurity CoP. These findings provide practical implications for CoP African members and a steppingstone on what to consider prior to implementing an African CoP for addressing cybersecurity challenges and concerns.

Abstract. Many universities are using lecture recording technology to expand the reach of their teaching programs, and to continue instruction when face to face lectures are not possi- ble. Increasingly, high-resolution 4K cameras are used, since they allow for easy reading of board/screen context. Unfortunately, while 4K cameras are now quite affordable, the back- end computing infrastructure to process and distribute a multitude of recorded 4K streams can be costly. Furthermore, the bandwidth requirements for a 4K stream are exorbitant - running to over 2GB for a 45-60 minute lecture. These factors mitigate against the use of such technology in a low-resource environment, and motivated our investigation into meth- ods to reduce resource requirements for both the institution and students. We describe the design and implementation of a low resource 4K lecture recording solution, which addresses these problems through a computationally efficient video processing pipeline. The pipeline consists of a front-end, which segments presenter motion and writing/board surfaces from the stream and a back-end, which serves as a virtual cinematographer (VC), combining this contextual information to draw attention to the lecturer and relevant content. The bandwidth saving is realized by defining a smaller fixed-size, context-sensitive ‘cropping window’ and generating a new video from the crop regions. The front-end utilises compu- tationally cheap temporal frame differencing at its core: this does not require expensive GPU hardware and also limits the memory required for processing. The VC receives a small set of motion/content bounding boxes and applies established framing heuristics to determine which region to extract from the full 4K frame. Performance results coupled to a user survey show that the system is fit for purpose: it is able to produce good presenter framing/context, over a range of challenging lecture venue layouts and lighting conditions within a time that is acceptable for lecture video processing.

Abstract. This paper shows how it is at least in principle possible to detect impostor nodes in wireless sensor networks with a quite simplistic detection algorithm by purely statistical means and merely from external observation without any knowledge of the impostor’s inter- nal composition. This method, however, requires considerable volumes of internal memory for any WSN node on which such detection algorithms are supposed to be implemented.
Keywords: Wireless Sensor Networks · WSN Reliability · WSN Integrity · WSN Safety and Security · WSN Trustworthiness · Node Replication Attacks · Intrusion Detection · Probabilistic Finite Automata · Randomised Detection Algorithm · Detection Accuracy · Memory Requirements

Abstract. Convolutional neural networks have proven to be very powerful for image classification problems, but still has its shortcomings in the presence of non-ideal data. Recently, facial recognition has become popular with usages such as surveillance and automatic tagging of individuals on social media sites. This paper explores a facial recognition solution that utilizes a feature masking strategy focused on facial landmarks with the goal of developing a solution capable of facial recognition in the presence of occlusions. The main driving factor behind this paper is based on the idea that the most commonly found occlusions in the wild are found in the regions of the facial landmarks and that these landmarks play a crucial role during the recognition process. It is found that using a masking strategy based on facial landmarks can be beneficial if the network is trained adequately and the dataset contains mostly well aligned faces, offering improved performance in comparison to using an arbitrary grid layout for all the tested occlusions. Furthermore, it is discovered that masks are not precise at removing the targeted features, causing the masking strategy to also harm recognition process in some cases by accidentally removing critical features.

Abstract. One of the most challenging problems faced by ecologists and other biological re- searchers today is to analyze the massive amounts of data being collected by advanced monitoring systems like camera traps, wireless sensor networks, high-frequency radio track- ers, global positioning systems, and satellite tracking systems being used today. It has become expensive, laborious, and time-consuming to analyze this huge data using man- ual and traditional statistical techniques. Recent developments in the deep learning field are showing promising results towards automating the analysis of these extremely large datasets. The primary objective of this study was to test the capabilities of the state-of- the-art deep learning architectures to detect birds in the webcam captured images. A total of 10592 images were collected for this study from the Cornell Lab of Ornithology live stream feeds situated in six unique locations in United States, Ecuador, New Zealand, and Panama. To achieve the main objective of the study, we studied and evaluated two con- volutional neural network object detection meta-architectures, single-shot detector (SSD) and Faster R-CNN in combination with MobileNet-V2, ResNet50, ResNet101, ResNet152, and Inception ResNet-V2 feature extractors. Through transfer learning, all the models were initialized using weights pre-trained on the MS COCO (Microsoft Common Objects in Context) dataset provided by TensorFlow 2 object detection API. The Faster R-CNN model coupled with ResNet152 outperformed all other models with a mean average preci- sion of 92.3%. However, the SSD model with the MobileNet-V2 feature extraction network achieved the lowest inference time (110ms) and the smallest memory capacity (30.5MB) compared to its counterparts. The outstanding results achieved in this study confirm that deep learning-based algorithms are capable of detecting birds of different sizes in differ- ent environments and the best model could potentially help ecologists in monitoring and identifying birds from other species.

Abstract. In this work, we consider the problem of procedural content generation for video game levels. Prior approaches have relied on evolutionary search (ES) methods capable of gener- ating diverse levels, but this generation procedure is slow, which is problematic in real-time settings. Reinforcement learning (RL) has also been proposed to tackle the same problem, and while level generation is fast, training time can be prohibitively expensive. We propose a framework to tackle the procedural content generation problem that combines the best of ES and RL. In particular, our approach first uses ES to generate a sequence of levels evolved over time, and then uses behaviour cloning to distil these levels into a policy, which can then be queried to produce new levels quickly. We apply our approach to a maze game and Super Mario Bros, with our results indicating that our approach does in fact decrease the time required for level generation, especially when an increasing number of valid levels are required.

Abstract. Crowd Density Estimation (CDE) can be used ensure safety of crowds by preventing stampedes or reducing spread of disease which was made urgent with the rise of Covid-19. CDE a challenging problem due to problems such as occlusion and massive scale varia- tions. This research looks to create, evaluate and compare different approaches to crowd counting focusing on the ability for dilated convolution to extract scale-invariant contex- tual information. In this work we build and train three different model architectures: a Convolutional Neural Network (CNN) without dilation, a CNN with dilation to capture context and a CNN with an Atrous Spatial Pyramid Pooling (ASPP) layer to capture scale-invariant contextual features. We train each architecture multiple times to ensure statistical significance and evaluate them using the Mean Squared Error (MSE), Mean Average Error (MAE) and Grid Average Mean Absolute Error (GAME) on the Shang- haiTech and UCF CC 50 datasets. Comparing the results between approaches we find that applying dilated convolution to more sparse crowd images with little scale variations does not make a significant difference but, on highly congested crowd images, dilated con- volutions are more resilient to occlusion and perform better. Furthermore, we find that adding an ASPP layer improves performance in the case when there are significant differ- ences in the scale of objects within the crowds. The code for this research is available at https://github.com/ThishenP/crowd-density.

Abstract. Given the vast amount of publicly available CCTV surveillance and the capabilities of modern computer vision algorithms, the task of automatic anomaly detection is due to be solved in the near future. A solution that is competent over the large problem domain requires a certain level of sophistication such that it can replicate the contextual understanding of a human monitor. It is hypothesised that a single approach to anomaly detection can not be expected to perform both low-level and high-level monitoring of video frames which is required for robust anomaly detection. This paper proposes a solution to the anomaly detection problem in the form of a consensus framework that combines inputs from three sources to provide a final verdict on the perceived degree of anomaly contained in a video. The first approach, later introduced as the base model, is an implementation of previous work in anomaly detection that is specifically chosen for its emphasis on the learning of high-level context. The second and third are novel anomaly detection heuristics that operate on a per-frame basis i.e., with no regard for high-level context. The paper concludes with an evaluation and analysis of the three approaches and a discussion of the merit of a consensus framework. A final AUC of 0.7156 is achieved on the UCF Crime dataset; however, this result is not attributable to the consensus framework.

Abstract. Methods such as the h-index and the journal impact factor are commonly used by the scientific community to quantify the quality or impact of research output. These methods rely primarily on citation frequency without taking the context of citations into consideration. Furthermore, these methods weigh each citation equally ignoring valuable citation characteristics, such as citation intent and sentiment. The correct classification of citation intents and sentiments can therefore be used to further improve scientometric impact metrics.
In this paper we evaluate BERT for intent and sentiment classification of in-text ci- tations of articles contained in the database of the Association for Computing Machinery (ACM) library. We analyse various BERT models which are fine-tuned with appropriately labelled datasets for citation sentiment classification and citation intent classification.
Our results show that BERT can be used effectively to classify in-text citations. We also find that shorter citation context ranges can significantly improve their classification. Lastly, we also evaluate these models with a manually annotated test dataset for sentiment classification and find that BERT-cased and SciBERT-cased perform the best.

Abstract. Establishing intelligent crop management techniques for preserving the soil, while providing next-generational food supply for an increasing population is critical. Nitrogen fertilizer is used in current farming practice as a way of encouraging crop development; however, its excessive use is found to have disastrous and long-lasting effects on the environment. This can be reduced through the optimization of fertilizer application strategies. In this work, we apply a set of reinforcement learning algorithms – the DQN, Double DQN, Dueling DDQN, and PPO – to learn novel strategies for reducing this application in a simulated crop growth setting. We provide an analysis of each agent’s ability and show that the Dueling DDQN agent can learn favourable strategies for minimizing nitrogen fertilizer application amounts, while maintaining a sufficient yield comparable to standard farming practice.

Abstract. Existing user studies on how users use digital archives as information systems seldom focus on what influences users’ needs and expectations. Similarly, not much is known about how the low resource context influences users’ needs. What users expect from searching and other related functionalities is rarely addressed in the cultural heritage and historical digital archives. These gaps unveil the mismatch between users’ needs (and expectations) and deployed technologies in the low resource context. As a result, delivering novel services through these digital archives is impossible because of the gap between design and reality. Users in the low resource environment are thus constrained to use whatever functionalities are available. This paper presents the empirical result of a user study. We determined the study’s sample framing using the future determination analysis technique. This analysis also guided the scoping of the study’s survey. The study foregrounds the need to adapt to users’ ever-changing expectations by understanding their needs. This is critical for a better system design that meets users’ expectations. A key finding is that users strongly prefer simple search functionalities in low resource environments. Regardless, they would prefer to use advanced features if given the opportunity. However, the expertise (and sometimes funding) needed to satisfy this desire is scarce. The surveyed users are only end-users without the expertise to innovate and build digital archives to meet their needs. This dearth of “resource(s)” was found to be characteristic of the experience of low resource (or resource-poor) settings like South Africa.

Abstract. Chronic diseases account for 71% of mortality across the globe. Health systems in both developed and developing countries are constrained in their ability to deal with the increasing incidence of chronic diseases due to the global shortage of healthcare professionals. Consequently, health social network sites (HSNs) are increasingly being used by patients and caregivers to search for health information and provide social support to one another. This paper presents the results of a systematic literature review (SLR) aimed at exploring the benefits and drawbacks of HSNs. Thirty-four research papers retrieved from five electronic databases were analyzed using specific inclusion/exclusion and quality assessment criteria. The results showed that all the 34 papers included in the SLR were published by authors from developed countries. Using thematic analysis, the benefits identified were classified broadly as (i) provision of health information and (ii) social support. The drawbacks include the dissemination of health misinformation, feeling of marginalization by healthcare professionals and concerns over privacy and confidentiality of health information. The research results highlight a need for studies that focus on the use of HSNs by researchers from developing countries and create a unique opportunity for African researchers to leverage the continuing growth in the number of active social media users to address the shortage of healthcare professionals in the continent.

Abstract. Knowledge is a strategic asset and organizations that have operational knowledge management (KM) systems, possess competitive advantage over their rivals. However, harnessing tacit knowledge and converting it to a form that is accessible and usable by all, is a great challenge. Tacit knowledge sharing is dependent on the will of the employee to share knowledge and the purpose of this research is to investigate the role played by culture in either hindering or enabling knowledge sharing. Using a single case study of an international multilateral organization, this research study investigated the effect that organizational culture has on KM. Data was collected through a questionnaire that had a combination of guided and open-ended questions. More than two-thirds of the employees of the case study organization participated in the research study. Data analysis was done in two ways. First, numerical data were analyzed using tallies and percentages for easier interpretation. Secondly, narrative analysis was used to analyze data from open-ended questions. It was established that some traits of organizational culture such as bureaucracy, being process oriented and being risk-averse hindered knowledge sharing in KM. Descriptors such as welcoming, transparent and teamwork were aspects that positively impacted KM. By understanding the impact of culture on knowledge sharing in KM, organizations may address potential barriers to KM and optimize the value of their KM effort.

Abstract. While much effort has been expended on understanding the adoption and implementation of big data analytics in organisations, less effort, comparatively speaking, has been put into investigating the business value that can be derived from such investments. Recent research on the resources and capabilities required to leverage big data analytics value offers promise in this regard. The purpose of this research study is to describe big data analytics capabilities required to create business value that arises from the work practice level. This study was a single case qualitative semi-structured interview research study. The study found that the functions using big data analytics at the work- practice level included marketing, customer relationship management, and product development. Capabilities identified include strategic alignment, human expertise, technology, culture, investment and time, and governance. The work practices enacted included inductive, deductive and abductive approaches, as well as algorithmic and human-based intelligence. Product innovation, market penetration, customer satisfaction and revenue growth represented the business value accrued to the organisation.

Abstract. On the 4th of January 2021, WhatsApp proposed a policy update that required its users to consent to have some of their data on the application linked to Facebook to boost the commercialisation of their users’ data for their insights. The move received widespread attention and criticism from some users, media, regulators, and organisations that have an interest in user data protection. The pervasiveness of WhatsApp meant that close to 2 billion users around the world were likely to be impacted by the policy. The event also caused some further debate about the ethical standards of individuals’ privacy and data security online. This qualitative interpretive case study aims to explore what WhatsApp users based in South Africa knew about the policy, how they perceived it, and how their perceptions influenced their reactions to the policy and WhatsApp usage in general. The results suggest that users who were studied accepted the policy out of necessity, not choice, citing WhatsApp's omnipresence and lack of similarly ubiquitous alternatives as their main reasons for staying with the service.

Abstract. Healthcare firms need to respond faster to the rapidly changing threat landscape. Risks to patient privacy and safety are increasing due to recent cyber-attacks. Healthcare firms are lagging in building cybersecurity capabilities prescribed by best practice approaches. The purpose of this case study is to identify the barriers to building a dynamic cybersecurity capability within a South African healthcare software services firm. The firm is a major provider of cloud-based software as a service (SaaS) solutions to medical practitioners and hospitals. The study used interviews and document analysis as primary data collection methods. Thematic analysis guided by a dynamic capability perspective was used to identify the internal and external barriers that could impede building a dynamic cybersecurity capability at a healthcare software services firm. The research recommends interventions to address cybersecurity barriers in healthcare software services firms.

Abstract. During the COVID-19 pandemic and mandated global lockdowns, people and busi- nesses started the extensive use of video-conferencing applications for staying connected. This surge in demand and the usability of video-conferencing services has been severely overlooked in developing countries like South Africa, where one-third of adults rely on mo- bile devices to access the internet, and the per-gigabyte data cost is among the highest in Africa. Considering these numbers, we conduct a two-pronged study where 1) we measure data consumption of different Android apps through data measurement experiments and 2) we conduct interviews and usability assessments with bandwidth-constrained users to bet- ter understand the usability and Quality of Experience (QoE) of mobile video-conferencing apps. Usability is the degree to which specified users can use a product to achieve specified goals. In contrast, QoE measures the subjective perception of the quality of an application and the level of delight or annoyance with a service. The key benefit of this study will be to inform organisations that seek to be inclusive about these tools’ relative usability by letting them know about the factors influencing users’ QoE.

Abstract. Anecdotal evidence suggested that South African Small to Medium Enterprises (SME) who have access to Artificial Intelligence (AI) tools as part of their enterprise resource planning software are not adopting these tools. This was seen as a problem because the SME sector forms the foundation for economic growth and the adoption of AI in this sector could enhance its ability to compete on a global stage. Hence the purpose of this research is to understand this lack of adoption. This qualitative study follows an interpretive philosophy and an inductive approach. Seven medium sized companies were selected across a variety of industry sectors and executives from each company were interviewed. The findings reveal that even though the participants generally have a clear understanding of the benefits of AI adoption and can articulate use cases, there are inhibiting factors preventing adoption. Primary among these inhibiting factors is the fear of losing control of critical business processes to a machine-based algorithm and the perceived lack of IT maturity to adopt and manage these AI tools. The value of these findings is that they provide an understanding of the barriers to AI adoption as well us highlighting the South African characteristic of reliance on informal networks to guide adoption decisions.

Abstract. Organisations need to be able to adopt AI successfully, but also responsibly. This requirement is not trivial, as AI can deliver real value to adopters. However, can also result in serious impacts on humans. AI’s technical capabilities make AI powerful, still the implementation of AI in organisations is not limited to the technical elements and requires a more holistic approach. An AI implementation within an organisation is a socio-technical system, with the interplay between social and technical components. When AI makes decisions that impact people, the socio considerations in AI adoption frame- works are paramount. Although technical adoption challenges are well researched and can overlap with aspects associated with traditional IT implementations, artificial intelli- gence adoption often faces additional social implication. This study focuses on these social challenges, which is a problem frequently experienced by many organisations. The study investigates how an organisation can increase adoption of AI as part of its quest to become more data-driven. This study was conducted at an automotive manufacturer’s analytics competence centre, located in South Africa. This paper describes the first iteration of a larger research effort that follows the design science research methodology. A socio-specific artificial intelligence adoption framework was created and can be used by organisations to help them succeed with their AI adoption initiatives in a responsible manner.