22505 research outputs found
Sort by
It Hurts to Become
No full textUsing the medium of oil painting, this thesis will investigate the ways that memories of my past inform and distort my perception of reality. Using domestic spaces as a framework in combination with my own body as the subject matter, I am creating compositions that redefine the meaning of home as a concept. Rather than a physical place of comfort, home is a cemetery for past selves that I revisit in my mind and grieve, no matter the time that has passed. By expressing this internal decay through physically scraping and sludging paint onto large scale surfaces, I aim to pull my audience into my world of isolation that unravels memories of my past and the cyclical ways they entrap me now, as I grow up
Analyzing The Hidden Impact Of Work Stress On Mental Health
No full textOccupational stress and deteriorating mental health constitute an increasing issue in the global community, and the problem of how job demands, personal resources, and cognitive processes interrelate to determine employee well-being is hardly known. This dissertation will study these dynamics by incorporating theory-driven Structural Equation Modeling (SEM) with data-driven Machine Learning (ML) methods that offer explanatory and predictive information. The study examines the influences of job demands, personal resources, effort–reward imbalance (ERI), emotional labor, cognitive appraisal, and social support on the stress and mental health outcomes of a sample of 5,000 employees working in various industries and regions of the world. The strong psychometric validity was confirmed by a comprehensive measurement model in which all the factor loadings were greater than 0.50, satisfactory reliability, and reasonable discriminant validity. Patterns of correlations demonstrated theoretically consistent correlations between job demands, cognitive appraisal, resources, and well-being measures. The structural model showed a very good fit and presented a number of interesting significant pathways: job demands, ERI/overcommitment, emotional labor, and conservation of resources were found to increase stress, whereas job resources, personal resources, and personal environment fit decreased it significantly. Stress had a serious, negative impact on cognitive appraisal (β = -.55) and a more moderate, direct, negative impact on mental health ( ( β = -.07). Notably, the immense indirect effect was significant when the researchers discovered that the negative impact of stress on mental health was reflected in the fact that it weakens the capacity of the employees to make an adaptive assessment of work demands. The results were that the cognitive appraisal had a significant beneficial effect on mental health (β = 0.26) although social support was found as a significant moderator that buffered the effect of stress on mental health to a significant degree ( β = -0.09). All these results are solid evidence in favor of Job Demands Resources and Effort Reward Imbalance models. Five machine learning models, including Decision Tree, Random Forest, K-Nearest Neighbors, Support Vector Machine, and Gradient Boosting, were trained to predict mental health outcomes in order to supplement the SEM analysis. The Support Vector Machine and the Gradient Boosting models had the best predictive results (R2 = 0.52, RMSE = 0.57), which indicates that it is possible to develop data-driven methods to estimate mental health risks. In all cases, cognitive appraisal, job demands, emotional labor, social support, self-determination, and access to mental health resources all proved to be the strongest predictors. These findings demonstrate a high similarity between causal pathways based on SEM and causal pathways based on ML, which confirms the validity of the conclusions reached in the study. In general, the dissertation has a theoretical value because it explains stress as a mediator between work conditions and mental health, and also a practical value in terms of the identification of crucial targets in the workplace with the purpose of interventions. The inclusion of SEM and ML also proves the importance of integrating both explanatory and predictive analytics in the research of organizations. The results provide a basis on which to create scalable survey-based mental health assessment instruments that could help detect early and intervene in the right place to prevent mental health issues in the workplace
Predicting Mental Health Risk in Remote Workers: A Machine Learning Approach
No full textThis dissertation explores the application of machine learning in mental health risk prediction of remote workers, which is increasingly becoming a significant issue, with the practise of flexible working reshaping organisational practises. It is based on the theoretical framework that has been used previously such as Job Demands Resources model and Stress-Strain model which emphasise the role of workload, support systems and personal resources in determining the well-being of employees. The context of the study indicates the growing rate of remote and hybrid employment, and the associated increase in the number of issues associated with stress and isolation, as well as mental pressure. The core research questions were considered to be the factors that affect mental health risk in remote workers, the effectiveness of different machine learning models to forecast the risk of these factors, and the predictive insights which can benefit organisational well-being strategies. A virtual survey dataset which included 5 000 samples of records of remote, hybrid, and onsite workers was employed to guarantee that the ethical consideration is still feasible but also record realistic patterns within the workplace. That analysis included cleaning, and preprocessing, encoding, and feature engineering data to get the dataset ready to modelling. Various algorithms of machine learning were applied, which are Logistic Regression, Decision Tree, Random Forest, Gradient Boosting, XGBoost, and Support Vector Machines. Accuracy, precision, recall, F1-score, and ROC-AUC were used as the evaluation models to identify the model that is best performing in terms of classifying mental health risk. The results showed that all of the models were very effective and consistent in recognising individuals that were identified as being at risk of mental health with a number of models having high recall and F1-scores. Such variables as the level of stress, the quality of sleep, social isolation, and satisfaction with remote work turned out to be significant predictors of risk. These findings support the theoretical knowledge that the psychological well being of the remote setting is influenced by both personal and organisational variables. The research has concluded that machine learning can offer a working solution to the early detection of employees who might need some support in order to apply more proactive to well-being measures in organisations. This and other patterns present in this synthetic environment should be tested in future studies through the use of real-world data or longitudinal data, which ii the dissertation recommends using. It also proposes the use of qualitative data to improve the contextual knowledge. To the organisational practise, the findings promote the incorporation of both data-driven well-being monitoring solutions and more conventional HR approaches to enhance the utilisation of remote and hybrid work-based employees
Unmasking Corruption And Bribery Using Predictive Analytics
No full textThe initiative (Unmasking Corruption and Bribery Using Predictive Analytics) aims to identify employees in government sectors who commit corruption and bribery by addressing issues such as the lack of transparency, fairness, equality, and public trust, as well as weakened loyalty and poor reputation. This project focuses on data analytics and predictive modeling to detect suspicious or high-risk employee behaviors related to corruption and bribery. The main objective is to reduce these unethical activities in government sectors, strengthen integrity, and promote transparency and public trust through predictive capabilities. It also seeks to raise awareness about how serious and harmful these crimes are. Since such activities are often committed in secret, it is crucial to monitor employees carefully. Data analysis and prediction using machine learning can greatly simplify the process of identifying individuals involved in these crimes. This strategy aims to eliminate such behaviors, promote honesty, and improve public trust, which will positively affect the state. It emphasizes the importance of ethical values among employees in all government sectors and how they can create a more positive and fair work environment. The predictive model findings indicate that the XGBoost model (46) outperformed other machine learning methods in detecting suspicious employee behavior, achieving the highest recall of (94%) (27). The model accurately identified strong indicators of corruption and bribery risks, such as accepting bribes or showing unusual approval patterns, as atypical and anomalous behaviors. These results highlight the success of advanced machine learning techniques in revealing hidden risks that might remain unnoticed by traditional moni- toring systems. The research concludes that predictive analytics can be an effective approach for monitoring public officials and detecting misconduct. Implementing these strategies en- ables government agencies to reduce corruption allegations, restore public trust, and promote transparency and fairness. Moreover, the findings demonstrate how integrating data-driven methods into decision-making supports ethical governance and strengthens accountability within public institutions. Keywords: Corruption, Bribery, techniques, lack of transparency, fairness, public trust, government sectors, employees, machine learning, random Forest, XGboost, logistic regrassion, neural network, LSVM, correlation, chi-square, Mann-withney, boxplot, mahalanobis
DATA-DRIVEN CRIME PREDICTION: TOWARD SMARTER REDUCTION STRATEGIES
No full textThis study explores how machine learning and weather data can be used for the prediction of crime more accurately in the city of Seattle. Predictive policing is a method in law enforcement that uses data and computer algorithms to forecast the locations that crimes are likely to happen. Although many studies focused on using past crime data alone, this research also includes weather conditions like temperature, rainfall, and humidity, which may influence when and where crimes occur. Several machine learning models, including Random Forest (RF) and Support Vector Machines (SVM), were used to classify areas of the city into high-risk and low-risk zones. These models were trained using a large dataset that included historical crime reports and weather records. To make the results easier to understand and apply in the real world, GIS tools and heatmaps were used to show crime hotspots and patterns across different parts of the city. The results showed that weather does have an impact on crime rates, especially for crimes like assault or theft. The models performed better when weather data was included, which shows that environmental factors should be part of future crime prediction systems. However, some challenges such as potential bias in historical crime data and the complexity of some machine learning models were seen, which can be hard to explain to non-technical users. Overall, this study shows that combining machine learning, environmental factors, and spatial mapping can create stronger and more useful crime prediction tools. This approach can help police departments plan patrols more effectively and use their resources in smarter and fairer ways
Detecting Fraud in Police Reports Using Machine Learning and Natural Language Processing
No full textThe paper explores how statistical analysis and machine learning can be used to identify the fraud patterns in the police reports. The study aims at establishing the most important predictive factors and indicators distinguishing fraudulent and valid cases with the use of structured data of police databases. The work was done in the background of the increase in financial fraud instances and the rising necessity of the introduction of automated detection systems in police departments. Police reports of the pastwere mined down to data and analyzed on SPSS 1, to carry out statistical operations. The sample was structured data which covered the financial transactions, demographic and behavioral pattern. It was based on the use of descriptive statistics, chi-square, ANOVA, and regression analysis as well as machine learning predictive models. Itwas found that financial indicators, especially, the amount and frequency of transactions, proved to be the most important predictors of fraud. Categorical variables such as transaction method and geographical location demonstrated that they were strongly associated with the fraud activities. The best classification level was accepted to be 0.6 probability where F1-score was at 0.8612 with a precision of 80.5% and recall of 92.7%. The paper finds that conventional statistical tools and machine learning present a solid framework of detecting fraud within the law enforcement setting. The results indicate that structured information itself has moderate predictive capabilities, saying that their functionality could be improved significantly by combining it with unstructured source of data. As a practice recommendation, the study suggests using automated monitoring protocols that target the specified key indicators and optimizing the best probability level to designate cases. Future studies are advised to examine how to combine the use of Natural Language Processing to analyze unstructured reports text in a way to assess the temporal variability of fraudulent behavior as well as to involve cross-jurisdictional comparative research to establish more empirically generalized detection frameworks
Predictive analysis of residential property in urban areas
No full textThis paper aims to determine the predictive modelling of residential property prices in urban Scotland with the help of an integrated framework in which machine learning methods are applied in combination with detailed socio-economic, health, housing, and geographic indicators. Conventional valuation methods tend to be based on the concept of few structural variables and ignore the effect of multidimensional variables in determining spatial variation in housing markets. To cope with this, the study uses Linear Regression, random forest, Multi-layer perceptron, and XGBoost models, which are assisted by a broad range of feature engineering, outlier management, and data preprocessing. Compared with any other model that was tested, XGBoost performed the best, and its RMSE was around £62,846 with an R2 of 0.739 on the log-transformed target. The model behaviour was analyzed using AI, especially SHAP, to determine the main aspects influencing the price change. The findings show that socio-economic deprivation, health outcomes, accessibility to services, and housing typologies have a great impact on the value of local property. The findings indicate the need to have multidimensional data integration and open machine learning methods in the study of urban housing markets. The paper finds that more equitable and evidence-based decision making could be supported through the use of advanced predictive models in planning, housing policy and resources allocation across the urban areas of Scotland
Segment-Level Machine Learning for Detecting Partial Deepfake Audio: An RNN–SVM Hybrid Approach For Real-World Adversarial Environments
No full textThis study investigates the detection of real, fully fake, and partially fake (PF) audio using classical machine-learning models as well as a segment-level analysis model. Unlike most existing research, which focuses solely on binary real-vs-fake classification, this work introduces a three-class detection framework and constructs realistic PF samples by inserting short synthetic speech segments into otherwise genuine audio recordings. Its method combines MFCC and spectral feature engineering,Wav2Vec2 embeddings, controlled PF synthesis and various models such as SVM, Random Forest, XGBoost and an attention based RNN. Segment level windowing allows fine-grained study of transition of time and breaks of manipulation. The findings reveal that the classical models are very accurate with fully fake audio- attributed to the occurrence of major artefacts in the world- but not with the PF samples, with short manipulations that are produced as acoustic melodies of real speech. The XGBoost had the highest overall performance, and SVM had the highest PF recall of all classical baselines. Predictions at the segment level also indicated boundary instability and demonstrated the drawbacks of MFCC features to predict short-scale transitions. These results reveal a severe security threat: PF audio can go around traditional detectors with only some necessary changes to key phrases. The findings of the study suggest that although classical pipelines are still useful in the context of full deepfakes, temporal structures, multimodal signals, and adversarial resistance are the key elements of robust PF detection, and future research should focus on implementing networks in real-time and using larger, more diverse datasets
Integrating Climate and Geospatial Features into Machine Learning Models
No full textThis thesis addresses a significant gap in real estate valuation models by investigating the economic impact of localized climate conditions and granular geospatial amenities. An abstract summarizes the following: - The main themes, ideas or areas of theory being investigated: This research investigates the integration of localized climate conditions and granular geospatial amenities into machine learning (ML) frameworks for residential real estate valuation. - The background and context of the research: In dynamic urban markets like Dubai, traditional valuation models often rely on broad location labels and structural attributes, overlooking the tangible economic impact of environmental comfort and micro-climates in a hot-arid environment. - The questions which informed data collection: The study sought to determine if modern ML models outperform linear baselines, whether integrating granular geospatial and climate data improves predictive accuracy, and which specific environmental features act as significant price determinants. - The main method(s) used to collect data and the sample: A quantitative approach fused three open data streams: Dubai Land Department transactions (2022–2024), Open-Meteo climate archives, and OpenStreetMap geospatial layers. Three models—Multiple Linear Regression, Random Forest, and XGBoost—were trained and evaluated using a feature ablation study. - A summary of the answers to the research questions: Random Forest achieved the highest accuracy ( 0.845), significantly outperforming the linear baseline. The ablation study revealed that geospatial features provided a substantial predictive lift ( +0.043), while novel features like distance to coastline and average temperature were identified as top-tier predictors. - The conclusions formed from these results: The study concludes that localized climate and granular geospatial features are not minor amenities but essential, quantifiable drivers of residential value, critical for accurate and resilient valuation in Dubai. - Recommendations for future research and for practice: Valuators are recommended to adopt ensemble ML models and standardly integrate geospatial data, while policymakers can use these findings to economically justify investments in climate-resilient urban infrastructure. Future research should expand this framework to include satellite-derived remote sensing data and different property types
AI-DRIVEN CYBER THREAT DETECTION
No full textBycreating an AI-driven method using deep learning and statistical analysis tools, this study seeks to fill important security holes in conventional intrusion detection systems. Current signature-based systems miss new and complex cyberattacks, which have significant financial and operational consequences for companies. The suggested approach detects unusual network activity in real-time by combining statistical analysis with long short-term memory networks (LSTMs), convolutional neural networks (CNNs), and statistical analysis. This study will create and test hybrid models that can identify both known and zero-day threats while reducing false positives using publicly accessible datasets like UNSW-NB15, CIC-IDS2017, and NSL-KDD. Expected results are a system for real-time threat detection that offers up to 75% quicker identification relative to conventional approaches and lowers false positive rates by about 80%. By tackling the rising difficulty of sophisticated cyber attacks in an ever more complicated digital environment, this study adds to the cybersecurity domain