8333 research outputs found
Sort by
The Impact of Colourism and Western Beauty Standards on the Self-Esteem of Women of Colour in Ireland
No full textThis study aimed to explore the relationship between how attitudes towards the components of colourism (self-concept, upward mobility, impression formation, and affiliation) and the endorsement of Western beauty standards affect the self-esteem of women of colour from various immigrant backgrounds in a predominately white country such as Ireland, addressing a critical gap within a global context as previous research suggests a need for intersectionality. A cross-sectional study design was employed (N=174) across diverse ethnic and immigrant backgrounds, using validated scales such as the Rosenberg Self-Esteem Scale, American Beauty Standards subscale and the In Group Colourism Scale. Multiple regression analysis revealed that the endorsement of Western beauty standards and upward mobility significantly predicted low self-esteem, explaining 54.4% of the variance. Generational and ethnic differences were analysed using two-way ANOVA’s, highlighting nuanced patterns within the descriptive statistics, as successive generations reported lower endorsement levels of Western beauty standards and attitudes towards colourism, but overall, there were no significant interactions between the variables. The findings of the results highlight the harmful effects of Western beauty standards and colourism on mental health and how prevalent this issue is, emphasising the need for culturally competent interventions, diverse media representation and educational initiatives to support self-acceptance and resilience among women of colour and their communities
Optimizing Resource Allocation in Cloud Computing using Machine Learning
No full textCloud computing enables on-demand access to shared computing resources, offering scalability, flexibility, and cost-efficiency for modern applications. However, efficient resource allocation remains a persistent challenge, as fluctuating workloads often lead to either over-provisioning, resulting in wasted resources, or under-provisioning, causing performance degradation. Traditional resource management strategies, such as rule-based and threshold-driven autoscaling, lack the predictive intelligence required to anticipate future demand accurately. This study addresses these limitations by implementing a cloud-native, predictive resource allocation framework using deep learning models integrated within Amazon Web Services (AWS). The proposed workflow involves collecting and preprocessing time-series data from over 1500 virtual machines (sourced from the GWA-T-13 Materna dataset), storing it in Amazon S3, and training forecasting models on EC2 instances using the Cloud9 IDE. Two models were implemented: a standard BiLSTM and an enhanced Attention-Centric BiLSTM Fusion model. The results demonstrate the superiority of the attention-based model, which achieved a lower Mean Squared Error (0.0043) and Root Mean Squared Error (0.0656), compared to the standard BiLSTM. The study successfully showcases how predictive modeling, when embedded in a secure and scalable cloud environment, can significantly improve resource utilization and support intelligent, cost-effective cloud infrastructure management
The relationship between demographic factors and familiarity and attitudes towards mental illness
No full textThis study investigates the relationship between demographic factors (gender, age, socio-economic status, education level) and familiarity and attitudes towards mental illness and borderline personality disorder in Ireland. Findings indicated that gender and familiarity were significant predictors, with females displaying more negative attitudes compared to males. Higher familiarity was unexpectedly linked to more negative attitudes. SES and education level were not statistically significant predictors, though individuals with higher SES showed more negative attitudes towards general mental illness, while those with lower SES held more negative views on borderline personality disorder. Age was also nonsignificant, though older adults showed a trend towards more negative attitudes. Limitations include a small sample size (N=92) and an over-representation of younger participants and individuals with low familiarity. These findings highlight the need for targeted stigma-reduction interventions and further research on females and individuals with high familiarity towards mental illness
Understanding Risks for Maternal Mortality in Rural Bangladesh Using XGBoost, Random Forest, and Decision Tree ML Models
No full textThis paper explores the application of machine learning models for predicting pregnancy risks, focusing on the performance comparison of XGBoost, Random Forest, and Decision Tree classifiers. The motivation behind this research stems from the critical need for early identification of high-risk pregnancies to improve maternal health outcomes. Using a dataset consisting of anonymous information from pregnant women in rural Bangladesh, this study implements feature scaling, standardization, and encoding to prepare the data. Both pre- and posthyperparameter tuning results are analysed, with additional focus on handling imbalanced data through the application of SMOTE (Synthetic Minority Oversampling Technique). The evaluation metrics include accuracy, precision, recall, F1-score, and ROC curves for each class. Key findings indicate that XGBoost outperforms the other models, particularly after hyperparameter tuning and SMOTE application, achieving an accuracy of 82%. The study emphasizes the importance of advanced machine learning techniques in healthcare, o↵ering significant implications for early and accurate prediction of pregnancy-related risks
AI-Powered System to Facilitate Personalized Adaptive Learning in Digital Transformation
No full textAs Large Language Models (LLMs) incorporate generative Artificial Intelligence (AI) and complex machine learning algorithms, they have proven to be highly effective in assisting human users with complex professional tasks through natural language interaction. However, in addition to their current capabilities, LLMs occasionally generate responses that contain factual inaccuracies, stemming from their dependence on the parametric knowledge they encapsulate. To avoid such inaccuracies, also known as hallucinations, people use domain-specific knowledge (expertise) to support LLMs in the corresponding task, but the necessary knowledge engineering process usually requires considerable manual effort from experts. In this paper, we developed an approach to leverage the collective strengths of multiple agents to automatically facilitate the knowledge engineering process and then use the learned knowledge and Retrieval Augmented Generation (RAG) pipelines to optimize the performance of LLMs in domain-specific tasks. Through this approach, we effectively build AI assistants based on particular customized knowledge to help students better carry out personalized adaptive learning in digital transformation. Our initial tests demonstrated that integrating a Knowledge Graph (KG) within a RAG framework significantly improved the quality of domain-specific outputs generated by the LLMs. The results also revealed performance fluctuations for LLMs across varying contexts, underscoring the critical need for domain-specific knowledge support to enhance AI-driven adaptive learning systems
Prediction of machining characteristics in coolant-assisted dry EDM of Inconel 625 and Titanium Grade 2 using Machine Learning
No full textDry Electro Discharge Machining (dry EDM) is an eco-friendly alternative to conventional EDM. Its adoption in industrial applications is limited due to the difficulty of stabilizing and the complexity of the process. So, identifying proper input parameters is fundamental for improving the efficiency of this process, which can be used for machining hard-to-machine alloys. In this work, four Machine Learning (ML) approaches describe the correlation of the dry EDM process inputs and outputs with distilled water as coolant for Inconel 625 and Titanium Grade 2. To compare the machinability of these two materials the Palatnik index Ψ was introduced that depends on physical properties. The prediction models based on Linear Regression (LR), Random Forest (RF), Support Vector Regression (SVR), and Artificial Neural Networks (ANN) receive the independent variables, pulse time, current, voltage, and gas pressure, to estimate the Material Removal Rate (MRR), the relative percentage wear of the working electrode (EW), the working electrode velocity (v), and the surface roughness parameters (Rz and Rsk). It was found that ANN outperforms other ML approaches in prediction of MRR, v, Rz and Rsk in case of prediction accuracy while the material and its Palatnik index is taken into account as an input. In addition, in the case of prediction of EWR, RF, ANN outperforms and other ML approaches considering all the prediction accuracy criteria. The average efficiency of the models in prediction of testing data which were not contributed to training stage according to the R-squared values for MRR, EW, and v were 0.6735, 0.7955, and 0.7739. The main aim of the research was to reduce the experimental time to identify optimal input parameters with respect to the desired output parameters using ML
Optimising window size of semantic of classification model for identification of in-text citations based on context and intent
No full textCitations in scientific literature act as channels for the sharing, transfer, and development of scientific knowledge. However, not all citations hold the same significance. Numerous taxonomies and machine learning models have been developed to analyze citations, but they often overlook the internal context of these citations. Moreover, it is worth noting that selecting the appropriate word embedding and classification models is crucial for achieving superior results. Word embeddings offer n-dimensional distributed representations of text, striving to capture the nuanced meanings of words. Deep learning-based word embedding techniques have garnered significant attention and found application in various Natural Language Processing (NLP) tasks, including text classification, sentiment analysis, and citation analysis. Current state-of-the-art techniques often use small datasets with fixed window sizes, resulting in the loss of contextual meaning. This study leverages two benchmark datasets encompassing a substantial volume of in-text citations to guide the selection of an optimal word embedding window size and classification approaches. A comparative analysis of various window sizes for in-text citations is conducted to identify crucial citations effectively. Additionally, Word2Vec embedding is employed in conjunction with deep learning models and machine learning models such as Convolutional Neural Networks (CNNs), Gated Recurrent Units (GRUs), Long Short-Term Memory (LSTM) networks, Support Vector Machines (SVM), Decision Trees, and Naive Bayes.The evaluation employs precision, recall, F1-score, and accuracy metrics for each combination of window sizes. The findings reveal that, particularly for lengthy in-text citations, larger citation windows are more adept at capturing the semantic essence of the references. Within the scope of this study, window sizes of 10 achieve superior accuracy and precision with both machine and deep learning models
Addressing Data Inequalities for Artificial Intelligence Technologies in Healthcare: Ways Forward for Policymaking
No full textThis policy paper was written based on a multi-method data collection [1], gathering information from academic journals, archival materials/grey literature and policy documents; participative insights (including 9 events); and 18 in-depth interviews with stakeholders working in civic society organizations, policy experts and Artificial Intelligence (AI) experts. The T&C and Privacy Policies of five popular AI engines were also reviewed.
The paper provides an overview of the promises and perils of generative AI technologies in healthcare. AI promises to change healthcare in previously unimaginable ways. Ireland advocates for the implementation and utilization of AI-based technologies to enhance public health and ensure that healthcare is more inclusive and accessible. However, previous research and experience clearly show that big tech companies dominate this space and their surveillance capitalist business models prioritize profit over social justice. This collaborative research, conducted in partnership with the Dublin Inner City Community Co-operative Society [2], delves into the power dynamics, political implications, and justice concerns surrounding data generation, utilization, and ownership in healthcare.
The paper gathers key recommendations for both the EU and its Member States, with a special focus on Ireland, given its significant role as the home to numerous big tech companies and a major controller of European users' data. The paper explores the tensions between the potential of generative AI and the dynamics of digital capitalism in Europe, and also looks at what impacts AI controversies are making in fostering data justice in healthcare. The recommendations explore what can be done to build moral (healthcare) data markets. The full list of recommendations is available in Section 6
Proactive Management of Delays in the French Railway Network: A Seasonal Machine Learning Based Approach
No full textTransportation planning is a critical component of effective urban growth, but traditional methods which are relying on manual procedures such as set schedules, fixed travel routes, and on paper ticketing infrastructure have difficulty keeping up with real-time data and changing passenger demands. In contrast, by examining train delays, their causes and utilising machine learning models such as Support Vector Regressor (SVR), Artificial Neural Network (ANN), Random Forest (RF), Decision Tree (DT), this research paper aims to improve the effectiveness of the system. The study makes use of hyperparameter tuning, exploratory data analysis and model evaluation metrics like MSE, RMSE, R2. Using a dataset with transit records from the French transportation network, the investigated models predict delays caused by various factors with very good accuracy. A Power BI dashboard was created to allow meaningful data exploration and it acted as a useful decision support tool for optimising delay. The results show that ANN was the most effective model with R-squared value 0.95 which is a great performance in anticipating delays. This research outcome demonstrates ANN’s strength and applicability for optimising proactive delay strategy in the challenging context of France railway system
An Enhanced Version of Data Classification based on Confidentiality for Cloud Security
No full textThe rapid proliferation of cloud technologies has motivated the organizations to store the data over cloud platforms since they offer high scalability and better performance in terms of software as a service. However, the increased deployment of cloud services has also increased the need for ensuring the protection of sensitive information stored in cloud servers. Privacy protection has become one of the critical aspects for various organizations that move their data to the cloud. Since the data can be of different types, the security requirements for data protection also vary. The crucial issue of securing data in cloud environments is addressed in this work by deploying an effective classification framework. This paper presents the design of a unique classification framework for securing the confidential data stored in the cloud. The classification model is developed in this work using the RandomForest (RF) classifier and the model is trained using the data features. The essential features are extracted using a hybrid CNN-LSTM model and a K-means SMOTE algorithm is used for addressing the class imbalance issues. Furthermore, the trained model is deployed into a Container as a Service (CaaS) environment and the deployed model is known as AUG-ConvoLSTM-RF. The model combines both data augmentation and Natural Language Processing (NLP) techniques for accurately classifying the data as confidential and non-confidential. The efficacy of the AUG-ConvoLSTM-RF model was experimentally evaluated and results show that the model exhibits an excellent classification accuracy of 84.36 % compared to other existing models