The concept of usability refers to a user's capability to interact with a system to fulfill goals in terms of task completion (effectiveness), time measurement (efficiency), and positive attitude (satisfaction). The strategy for usability evaluation in software systems usually involves questionnaires, user testing, and heuristics. Although these methods have been widely used due to several benefits, there are challenges related to time consumption and embedded bias. In response to these challenges, this work proposes a hybrid approach based on usability questionnaire answers and machine learning algorithms to predict usability scores. We describe three different experiments with features extracted from a Learning Management System. These features were applied in the Machine Learning algorithms Linear Regression, Decision Trees, Random Forest, and Neural Networks in three experiments. Random Forest produces the best performance of average mean square error and root mean square error among machine learning algorithms. The results are promising, though there are alternatives for improvements for better performance of the System Usability Scale and UseLearn scores prediction. This approach has potential as a reliable predictive tool for usability scores, which would help create software systems that better satisfy users' needs. / Master of Science / Instructors and students have used online platforms known as Learning Management Systems (LMSs) to improve learning and satisfaction. Students need to achieve their learning goals by interacting with these systems. To achieve these goals, usability evaluation involves ensuring that LMSs attain effectiveness (task completion), efficiency (time measurement), and satisfaction (positive attitude). Usability evaluation usually follows questionnaires, user testing of the LMS, and expert reviews. Although these methods are widely used due to several benefits, they face challenges related to trying these software systems multiple times until the system satisfies student needs and human subjectivity perception. To face these challenges, promote student engagement with the system, and create a better design in the LMS courses, we propose a hybrid approach based on data, questionnaire answers, and machine learning algorithms to predict usability scores. We evaluated this approach through a case study with data collected from undergraduate students at Virginia Tech. The results showed different advantages and drawbacks of machine learning performance. The approach contributes to the engineering and computing education field by providing a reliable predictive tool for usability scores to improve the student learning experience and the features of the LMS.

This metaverse and computer vision-powered application is designed to serve people with low vision or a visual impairment, ranging from adults to old age. Specifically, we hope to improve the situational awareness of users in a scene by narrating the visual content from their point of view. The user would be able to understand the information through auditory channels as the system would narrate the scene's description using speech technology. This could increase the accessibility of visual-spatial information for the users in a metaverse and later in the physical world. This solution is designed and developed considering the hypothesis that if we enable the narration of a scene's visual content, we can increase the understanding and access to that scene. This study paves the way for VR technology to be used as a training and exploration tool not limited to blind people in generic environments, but applicable to specific domains such as military, healthcare, or architecture and planning. We have run a user study and evaluated our hypothesis about which set of algorithms will perform better for a specific category of tasks - like search or survey - and evaluated the narration algorithms by the user's ratings of naturalness, correctness and satisfaction. The tasks and algorithms have been discussed in detail in the chapters of this thesis. / Master of Science / The solution is built using an object detection algorithm and virtual environments which run on the web browser using X3DOM. The solution would help improve situational awareness for normal people as well as for low vision individuals through speech. On a broader scale, we seek to contribute to accessibility solutions. We have designed four algorithms which will help user to understand the scene information through auditory channels as the system would narrate the scene's description using speech technology. The idea would increase the accessibility of visual-spatial information for the users in a metaverse and later in the physical world.

The rare-class data classification problem is a common one. It occurs when, in a dataset, the class of interest is far outweighed by other classes, thus making it difficult to classify using typical classification algorithms. These types of problems are found quite often in biological datasets, where data can be sparse and the class of interest has few representatives. A variety of solutions to this problem exist with varying degrees of success. In this paper, we present our solution to the rare-class problem. This solution uses MetaCost, a cost-sensitive meta-classifier, that takes in a classification algorithm, training data, and a cost matrix. This cost matrix adjusts the learning of the classification algorithm to classify more of the rare-class data but is generally unknown for a given dataset and classifier. Our method uses three different types of optimization techniques (greedy, simulated annealing, genetic algorithm) to determine this optimal cost matrix. In this paper we will show how this method can improve upon classification in a large amount of datasets, achieving better results along a variety of metrics. We will show how it can improve on different classification algorithms and do so better and more consistently than other rare-class learning techniques like oversampling and under-sampling. Overall our method is a robust and effective solution to the rare-class problem. / Ph. D.

Research across healthcare and organizational settings demonstrates the importance of social support to increase physical and mental well-being. However, the process of seeking social support is less well-understood than its outcomes. Specifically, research examining how people seek social support in natural settings is scarce. One natural setting increasingly used by people to seek support is the internet. In this online setting, people seek and provide social support verbally via social media platforms and messages. The present project seeks to further examine the nature of social support seeking in these online contexts by examining people’s language. This analysis includes discovering the common language features of social support seeking. By applying a data-driven content analysis approach, this research can examine the underlying themes present when seeking social support and build upon that insight to classify new instances of support seeking. These results would have important practical implications for occupational health. By identifying individuals who are seeking social support, future interventions will be able to take a more targeted approach in lending additional support to those individuals who have the greatest need. Subsequently, this application potentially provides the mental and physical health benefits of social support. Therefore, this research extends our knowledge of both the nature of support seeking and how to develop effective interventions. / M.S. / Research suggests that social support has important effects on our mental and physical health. To this point, though, the process of seeking social support has largely been neglected in research. Specifically, there hasn’t been much research on how social support is sought online. We know that people seek social support online by posting and messaging on social media. The present study seeks to examine the language of online support seeking—this way, we can understand what people tend to say when seeking support. The present study is concerned with the content of support seeking posts; by analyzing this content, we can understand themes that are prevalent in online support seeking. This allows us to better understand support seeking and, hopefully, better identify people in need of support. By identifying those people in need of support, we can ensure that their support needs are met and that they don’t suffer the health consequences related to a lack of social support. Therefore, this research extends our knowledge of social support seeking, both theoretically and practically.

Analytical solutions such as the Theis solution have historically been utilized to forecast changes in aquifer water levels resulting from human-driven withdrawals using pumping wells. This method, however, suffers from a number of disadvantages, such as long data acquisition times, model uncertainty, and trial-and-error calibrations. This study illustrated the effectiveness of alternate forecasting methods that utilized machine learning principles. The groundwater level dynamics of two sites located at the Virginia Eastern Shore were predicted using historical groundwater level below land surface (GWLBLS) data as the endogenous variable and local pumping data as the exogenous variable. Predicting the local pumping data from the GWLBLS values was also implemented, to not only enforce reliability of the model, but also to highlight the capability of verifying and enforcing permitted pumping data. The machine learning methods chosen for this study were the Random Forest and SARIMAX models. Historical datasets were divided into training/calibration and testing/validation sets, and the respective models were fit to the data. These calibrated models were then compared to the performance of the Theis solution. Across both study sites, the Random Forest performed best at forecasting groundwater level over time given the pumping data as an exogenous variable, with SARIMAX performing similarly to the Theis solution. The Theis solution, however, did perform well in terms of generalization ability (GA). / Master of Science / Groundwater is a vital resource for drinking, agriculture, and industry. In order to ensure aquifer health for future use, it is crucial to be able to forecast well water level in the midst of groundwater pumping. Currently, analytical solutions such as the Theis solution are utilized to predict water level over time, but data acquisition is time-consuming and many of the calibrations have to be based on trial-and-error. In this thesis, machine learning methods were explored as alternatives to the current analytical methods. The groundwater level dynamics of the two study sites, Oyster, Virginia and Temperanceville, Virginia, were used to calibrate corresponding machine learning models, called the Random Forest (RF) and SARIMAX models. While the Theis showed that it was the most adaptable model, the RF performed the best overall in terms of root mean square error and R2 scores, which were used as reliability metrics. This study provides a range of substitutes for the Theis solution that have the capability to perform better when calibrated on a site-by-site basis.

Due to the ever increasing supply of new Internet of Things (IoT) devices being added onto a network, it is vital secure the devices from incoming cyber threats. The manufacturing process of creating and developing a new IoT device allows many new companies to come out with their own device. These devices also increase the network risk because many IoT devices are created without proper security implementation. Utilizing traffic patterns as a method of device type detection will allow behavior identification using only Internet Protocol (IP) header information. The network traffic captured from 20 IoT devices belonging to 4 distinct types (IP camera, on/off switch, motion sensor, and temperature sensor) are generalized and used to identify new devices previously unseen on the network. Our results indicate some categories have patterns that are easier to generalize, while other categories are harder but we are still able recognize some unique characteristics. We also are able to deploy this in a test production network and adapted previous methods to handle streaming traffic and an additional noise categorization capable of identify non-IoT devices. The performance of our model is varied between classes, signifying that much future work has to be done to increase the classification score and overall usefulness. / Master of Science / IoT (Internet of Things) devices are an exploding field, with many devices being created, manufactured, and utilized per year. With the rise of so many internet capable devices, there is a risk that the devices may have vulnerabilities and exploits able to allow unauthorized users to access. While a problem for a consumer network, this is an increased problem in an enterprise network, since much of the information on the network is sensitive and should be kept confidential and private. While a ban of IoT devices on a network is able to solve this problem, with the rise of machine learning able to characterize and recognize patterns, a smarter approach can be created to distinguish when and which types of IoT devices enter the network. Previous attempts to identify IoT devices used signature schemes specific to a single device, but this paper aims to generalize traffic behaviors and identifying a device category rather than a specific IoT device to ensure future new devices can also be recognized. With device category identification in place on an internet network, smarter approaches can be implemented to ensure the devices remain secure while still able to be used.

No / Big data is big business. But having the data and the computational power to process it isn't nearly enough to produce meaningful results. Big Data, Data Mining, and Machine Learning: Value Creation for Business Leaders and Practitioners is a complete resource for technology and marketing executives looking to cut through the hype and produce real results that hit the bottom line. Providing an engaging, thorough overview of the current state of big data analytics and the growing trend toward high performance computing architectures, the book is a detail-driven look into how big data analytics can be leveraged to foster positive change and drive efficiency. With continued exponential growth in data and ever more competitive markets, businesses must adapt quickly to gain every competitive advantage available. Big data analytics can serve as the linchpin for initiatives that drive business, but only if the underlying technology and analysis is fully understood and appreciated by engaged stakeholders.

Yes / Twitter has become a fertile place for rumors, as information can spread to a large number of people immediately. Rumors can mislead public opinion, weaken social order, decrease the legitimacy of government, and lead to a significant threat to social stability. Therefore, timely detection and debunking rumor are urgently needed. In this work, we proposed an Attention-based Long-Short Term Memory (LSTM) network that uses tweet text with thirteen different linguistic and user features to distinguish rumor and non-rumor tweets. The performance of the proposed Attention-based LSTM model is compared with several conventional machine and deep learning models. The proposed Attention-based LSTM model achieved an F1-score of 0.88 in classifying rumor and non-rumor tweets, which is better than the state-of-the-art results. The proposed system can reduce the impact of rumors on society and weaken the loss of life, money, and build the firm trust of users with social media platforms.

The quality of an input sketch can affect performance of the computational algorithms. However, the quality of a sketch is not often considered when working with sketch tasks and automated sketch quality prediction has not been previously studied. This thesis presents quality prediction on the "Sketchy" dataset. The method presented here predicts a quality label rather than a zero to one quality metric. This thesis predicts an understandable label rather than a computer-generated quality metric with no human input. Previous tasks like sketch classification have used a transformer architecture to leverage the vector format of sketches. The architecture used in sketch classification was called Sketchformer. The Sketchformer was adopted and trained to predict quality labels of hand-drawn sketches. This Sketchformer architecture achieves 66% accuracy when predicting the 5-labels. The same transformer achieves up to 97% accuracy in a different experiment when combining the different labels into good versus bad (2-label) experiments. The sketchformer significantly outperforms the SVM baseline. The results of the experiments show that the transformer embedding space facilitates separation of 'good' sketch quality from 'bad' sketch quality with high accuracy. / Master of Science / If pictures are worth 1000 words, then sketches are worth a few hundred words. Sketches are easy to create using a pen and tablet. Objects in the sketches can be drawn many ways, depending on the talent of the creator and pose of the object. The quality of the sketches vary pretty drastically. When using sketches in computer vision tasks, the quality of a sketch can affect the performance of the computational algorithm. However, the quality of a sketch is not often considered when working with other sketch tasks. One common sketch task is called Sketch-Based Image Retrieval (SBIR). The input of this task is the sketch of an object/subject, and the model returns a matching image of the same object/subject. If the quality of the input sketch is bad, the output of this model will be poor. This thesis predicts the quality of sketches. The dataset used is called the "Sketchy" dataset, this dataset was originally used to study SBIR. However, the creators of the dataset provided quality labels for the sketches. This allows for quality prediction on this dataset, which has not previously been completed. There are 5 different labels assigned to sketches. One of the experiments completed for this thesis was predicting 1 of the 5 labels for each sketch. The other experiments for this thesis create good and bad labels by combining the 5 labels. The Sketchformer architecture created by Ribeiro et al. is used to run the experiments. The Sketchformer achieves 66% on the 5-label experiment and up to 97% on the good and bad (2-label) experiment. This transformer outperforms a Support Vector Machine baseline on this quality labels. The results of the experiments show that the transformer applied to this dataset is a valuable contribution by surpassing the baseline on multiple tasks. Additionally, accuracy values from these experiments are similar to values found in the corresponding image quality prediction task.

Augmented reality task guidance systems provide assistance for procedural tasks, which require a sequence of physical actions, by rendering virtual guidance visuals within the real-world environment. An example of such a task would be to secure two wood parts together, which could display guidance visuals indicating the user to pick up a drill and drill each screw. Current AR task guidance systems are limited in that they require AR system experts for use, require CAD models of real-world objects, or only function for limited types of tasks or environments. We propose a general-purpose AR task guidance approach and proof-of-concept system to generate guidance for tasks defined by natural language. Our approach allows an operator to take pictures of relevant objects and write task instructions for an end user, which are used by the system to determine where to place guidance visuals. Then, an end user can receive and follow guidance even if objects change location or environment. Guidance includes reusable visuals that display generic actions, such as our system's 3D hand animations. Our approach utilizes current vision-language machine learning models for text and image semantic understanding and object localization. We built a proof-of-concept system using our approach and tested its accuracy and usability in a user study. We found that all operators were able to generate clear guidance for tasks in an office room, and end users were able to follow the guidance visuals to complete the expected action 85.7% of the time without knowledge of their tasks. Participants rated that our system was easy to use to generate guidance visuals they expected. / Master of Science / Augmented Reality (AR) task guidance systems provide assistance for tasks by placing virtual guidance visuals on top of the real world through displays. An example of such a task would be to secure two wood parts together, which could display guidance visuals indicating the user to pick up a drill and drill each screw. Current AR task guidance systems are limited in that they require AR system experts for use, require detailed models of real-world objects, or only function for limited types of tasks or environments. We propose a new task guidance approach and built a system to generate guidance for tasks defined by written instructions. Our approach allows an operator to take pictures of relevant objects and write task instructions for an end user, which are used by the system to determine where to place digital visuals. Then, an end user can receive and follow guidance even if objects change location or environment. Guidance includes visuals that display generic actions, such as our system's 3D hand animations that mimic human hand actions. Our approach utilizes AI models for text and image understanding and object detection. We built a proof-of-concept system using our approach and tested its accuracy and usability in a user study. We found that all operators were able to generate clear guidance for tasks in an office room, and end users were able to follow the guidance visuals to complete the expected action 85.7% of the time without knowledge of the tasks. Participants rated that our system made it easy to write instructions and take pictures to create guidance visuals.