Miada Almasre’s research while affiliated with King Abdulaziz University and other places

Background The efficient performance of an Emergency Department (ED) relies heavily on an effective triage system that prioritizes patients based on the severity of their medical conditions. Traditional triage systems, including those using the Canadian Triage and Acuity Scale (CTAS), may involve subjective assessments by healthcare providers, leading to potential inconsistencies and delays in patient care. Objective This study aimed to evaluate six Machine Learning (ML) models K-Nearest Neighbors (KNN), Support Vector Machine (SCM), Decision Tree (DT), Random Forest (RF), Gaussian Naïve Bayes (GNB), and Light GBM (Light Gradient Boosting Machine) for triage prediction in the King Abdulaziz University Hospital using the CTAS framework. Methodology We followed three essential phases: data collection (7125 records of ED patients), data exploration and processing, and the development of machine learning predictive models for ED triage at King Abdulaziz University Hospital. Results and conclusion The overall predictive performance of CTAS was the highest using GNB = 0.984 accuracy. The CTAS-level model performance indicated that SVM, RF, and LGBM achieved the highest performance regarding the consistency of precision and recall values across all CTAS levels.

Criminals, including terrorists, may use Twitter to communicate and share their ideologies. They often employ multiple accounts for anonymity. While the other accounts hide their identities and use it for different purposes (e.g., communication with unknown criminals), they use one to write tweets revealing their beliefs and spreading evil thoughts (e.g., racism and bullying). Since these multiple accounts will not have the same contents, stakeholders cannot rely on the contents to detect various accounts belonging to the same person. Using stylometric features may help to detect these accounts as they depend on the writing style of a person rather than the content of the words and their meaning. In this paper, we build a model and use stylometric features that differ from the state-of-the-art, such as n-gram for part-of-speech tag, frequency of repeated characters, number of emojis, and more. These features distinguish our approach from existing methods. We evaluated different machine and deep learning classifiers to make comparisons. Our results highlight the effectiveness of our feature set, achieving a remarkable accuracy of 96%. Additionally, our findings indicate that machine learning classifiers exhibit superiority over their deep learning counterparts in the context of this study. Furthermore, we developed a comprehensive dashboard that offers users an in-depth analysis of different Twitter accounts and ranks their similarities. The dashboard serves as a valuable tool for gaining insights into the relationships and similarities among the analyzed Twitter accounts.

The increased use of Internet of Things (IoT) devices has led to greater threats to privacy and security. This has created a need for more effective cybersecurity applications. However, the effectiveness of these systems is often limited by the lack of comprehensive and balanced datasets. This research contributes to IoT security by tackling the challenges in dataset generation and providing a valuable resource for IoT security research. Our method involves creating a testbed, building the `Joint Dataset’, and developing an innovative tool. The tool consists of two modules: an Exploratory Data Analysis (EDA) module, and a Generator module. The Generator module uses a Conditional Generative Adversarial Network (CGAN) to address data imbalance and generate high-quality synthetic data that accurately represent real-world network traffic. To showcase the effectiveness of the tool, the proportion of imbalance reduction in the generated dataset was computed and benchmarked to the BOT-IOT dataset. The results demonstrated the robustness of synthetic data generation in creating balanced datasets.

The recent advancements in the fields of AI technology, generative AI, and Large Language Models (LLMs) have increased the potential of the deployment of such tools in educational environments, especially in contexts where student assessment fairness, quality, and automation are a priority. This study introduces an AI-enhanced evaluation tool that utilizes OpenAI’s GPT-4 and the recently released custom GPT feature to evaluate the typography designs of 25 students enrolled in the Visual Media diploma offered by King Abdulaziz University. A mixed methods approach is adopted to evaluate the performance of this tool against the rubric-based evaluations offered by two human evaluators, considering both grading and text feedback. The results indicate that there are statistically significant differences between the AI tool’s grading and feedback when compared to that of Evaluator 2; however, none is reported with Evaluator 1. The study presents a qualitative interpretation of the comprehensive feedback by the evaluator and reflects in further research in this area.

The Internet of Things (IoT) is an emerging technology that attracted considerable attention in the last decade to become one of the most researched topics in computer science studies. This research aims to develop a benchmark framework for a public multi-task IoT traffic analyzer tool that holistically extracts network traffic features from an IoT device in a smart home environment that researchers in various IoT industries can implement to collect information about IoT network behavior. A custom testbed with four IoT devices is created to collect real-time network traffic data based on seventeen comprehensive scenarios of these devices’ possible interactions. The output data is fed into the IoT traffic analyzer tool for both flow and packet levels analysis to extract all possible features. Such features are ultimately classified into five categories: IoT device type, IoT device behavior, Human interaction type, IoT behavior within the network, and Abnormal behavior. The tool is then evaluated by 20 users considering three variables: usefulness, accuracy of information being extracted, performance and usability. Users in three groups were highly satisfied with the interface and ease of use of the tool, with scores ranging from 90.5% to 93.8% and with an average score between 4.52 and 4.69 with a low standard deviation range, indicating that most of the data revolve around the mean

Recently, the problem of creating descriptive captions for images became a significant one. However, human languages’ expressivity had been among the challenges that hindered researchers from widely experimenting with creating linguistically rich captions for images. That motivated us to utilize advanced deep learning algorithms to generate captions for images. The researchers proposed an AI model utilizing deep learning and natural language processing algorithms, which has two main components, an image-feature extractor, and a story generator. The researchers trained the first component (image-feature extractor) of the model to predict object names in images. The second component (story-generator) was trained on a custom short descriptive sentence which considered short stories. So, the output from the first component (list of words) will be entered into the second component to generate stories on input images. Thus, when testing the model’s performance, a list of names will be entered from the first component so that the second generator arranges them and generates a short story from them. The proposed model developed could generate a short story expressive of an input image as shown by the results of a logical value used on the BLEU scale of 0.59, which further research is planned to improve.

Arabic Sign Language (ArSL) is similar to other sign languages in terms of the way it is gestured and interpreted and used as a medium of communication among the hearing-impaired and the communities in which they live in. Research investigating sensor utilization and natural user interfaces to facilitate ArSL recognition and interpretation, is lacking. Previous research has demonstrated that there is not a single classifier modeling approach that can be suitable for all hand gesture recognition tasks, therefore, this research investigated which combination of algorithms, set with different parameters used with a sensor device, produce higher ArSL recognition accuracy results in a gesture recognition system. This research proposed a dynamic prototype model (DPM) using Kinect as a sensor to recognize certain ArSL gestured dynamic words. The DPM used eleven predictive models of three algorithms (SVM, RF, KNN) based on different parameter settings. Research findings indicated that highest recognition accuracy rates for the dynamic words gestured were achieved by the SVM models, with linear kernel and cost parameter = 0.035.

The objective of this research was to recognize the hand gestures of Arabic Sign Language (ArSL) words using two depth sensors. The researchers developed a model to examine 143 signs gestured by 10 users for 5 ArSL words (the dataset). The sensors captured depth images of the upper human body, from which 235 angles (features) were extracted for each joint and between each pair of bones. The dataset was divided into a training set (109 observations) and a testing set (34 observations). The support vector machine (SVM) classifier was set using different parameters on the gestured words’ dataset to produce four SVM models, with linear kernel (SVMLD and SVMLT) and radial kernel (SVMRD and SVMRT) functions. The overall identification accuracy for the corresponding words in the training set for the SVMLD, SVMLT, SVMRD, and SVMRT models was 88.92%, 88.92%, 90.88%, and 90.884%, respectively. The accuracy from the testing set for SVMLD, SVMLT, SVMRD, and SVMRT was 97.059%, 97.059%, 94.118%, and 97.059%, respectively. Therefore, since the two kernels in the models were close in performance, it is far more efficient to use the less complex model (linear kernel) set with a default parameter.