8333 research outputs found
Sort by
Renewable power generation and weather conditions
No full textTraditional methods in power prediction like linear regression or even simple machine learning models have struggled to handle the complexities in time series data. This study explores the prediction of solar power generation using two machine learning models: Random Forest and Long Short-Term Memory (LSTM) networks using a data set obtained from two solar power plants in India. Records in the dataset cover 34 days in total, during which, there are the power generation record per 15 minutes and the weather data or the ambient temperature, module temperature, and irradiation. The main purpose to estimate the TOTAL_YIELD of solar plant in respect of weather and power generation characteristics. The first transforming process is the data pre-processing in which data is cleaned and converted to a supervised form With the help of a feature extractor, data is divided into training and testing sets. For model implementation, the Random Forest Regressor is employed to predict the total yield and the LSTM model to analyze time series data. The performance of both models is assessed using mean squared error (MSE), root mean squared error (RMSE), mean absolute error (MAE) and R² score. The outcomes reveal that the two models are reasonably accurate but the LSTM model provides better predictions with lesser error rates, which confirm the viability of time series forecasting. The final analysis is based on identifying the strengths of LSTM in terms of forecasting sequential data for renewable energy and, at the same time, the interpretability of Random Forest for features’ importance
Leveraging Advanced Machine Learning Models to Analyse Mental Health in the Era of Social Media and Digital Platforms
No full textSocial networks present both opportunities and challenges in adolescent mental health, acting as significant socio-environmental vulnerability factors. This study evaluates the effectiveness of social media-based detection methods in identifying early symptoms of emerging mental health issues among new students. The performance of these methods is assessed using key evaluation metrics such as precision, recall, F1-score, and AUC-ROC, comparing their efficacy against traditional intervention approaches. The findings highlight the scalability of social media for early detection and intervention while addressing ethical concerns related to privacy and consent. This research provides actionable insights for policymakers, mental health practitioners, and platform managers on the responsible integration of social networks into mental health care strategies. By contributing to the ongoing discourse on digital technology’s role in adolescent mental health, this study underscores the potential of social media as a proactive tool in mental health intervention
Standardised Versioning of Datasets: a FAIR–compliant Proposal
No full textThis paper presents a standardised dataset versioning framework for improved reusability, recognition and data version tracking, facilitating comparisons and informed decision-making for data usability and workflow integration. The framework adopts a software engineering-like data versioning nomenclature (“major.minor.patch”) and incorporates data schema principles to promote reproducibility and collaboration. To quantify changes in statistical properties over time, the concept of data drift metrics (d) is introduced. Three metrics (dP, dE,PCA, and dE,AE) based on unsupervised Machine Learning techniques (Principal Component Analysis and Autoencoders) are evaluated for dataset creation, update, and deletion. The optimal choice is the dE,PCA metric, combining PCA models with splines. It exhibits efficient computational time, with values below 50 for new dataset batches and values consistent with seasonal or trend variations. Major updates (i.e., values of 100) occur when scaling transformations are applied to over 30% of variables while efficiently handling information loss, yielding values close to 0. This metric achieved a favourable trade-off between interpretability, robustness against information loss, and computation time
Developing a QoS and Spatially aware scalable fog system with adaptive cache
No full textWith the expansion in the number of mobile systems integrated into IoT and fog networks, fog computing now includes scaling out and optimizing communication for mobile fog systems effectively. This research focuses on developing a QoS (Quality of Service) and spatially aware scalable fog system with an adaptive cache to enhance the efficiency and reliability of the system. In order to maintain a system that knows the approximate location of its devices even when they are constantly moving around. This is to optimize resource allocation, reduce latency, and improve the scheduling for these systems. Making use of existing ideas in fog computing, QoS provisioning, spatial awareness, and some caching techniques, a flat distributed network was developed which is also able to switch to a tiered network when offloading becomes required. The scheduler uses an adaptive cache to keep details of each fog Device in the configuration, it also is able to treat VMs on each fog device as independent machines while retaining QoS-aware properties such as Bandwidth and latency. Key performance metrics which were measured, analysed and compared against other possible more traditional fog systems without spatial awareness and adaptive caching were the energy and bandwidth usage and most importantly the savings in latency to show the new system’s efficiency. The application domain of choice is the autonomous car Industry, where latency-dependent decision-making could be crucial in averting disaster. The final results were able to show that keeping the system spatially aware was able to improve the QoS performance of the fog setup
A Comparative Analysis for Recognizing Emotions from Facial Expressions
No full textEmotions, categorized into anger, disgust, fear, gladness, neutrality, sadness, and surprise, significantly influence judgments and discussions on various issues. Deep learning, an artificial intelligence technique, can mimic the human brain’s data analysis to identify patterns for judgments. It uses networks to comprehend unsupervised, unstructured or unlabeled data, surpassing machine learning when dealing with large amounts of data. Unlike traditional programs, which examine data in a linear fashion, deep learning systems use a hierarchical function to handle data in a nonlinear fashion. For this research, the models developed for experimentation is a CNN model, a hybrid CNN-LSTM model, and VGG-16 model. The overall performed model was the hybrid CNN-LSTM model which gave an accuracy of 42% for a balanced dataset and 62% for unbalanced dataset. The model that performed best with a high accuracy was the CNN model that gave up to 62% in testing but gave a very high 99% accuracy during its training phase
Text-to-image generation using GAN
No full textThis research delves into the pursuit of generating realistic images from textual descriptions, a compelling yet challenging task within current AI systems. While existing technologies fall short of this goal, recent advancements in recurrent neural networks have demonstrated proficiency in learning discriminative text features. Additionally, deep convolutional generative adversarial networks (GANs) have shown promise in generating highly realistic images across specific categories like faces, album covers, and interiors. In this study, I introduce a novel deep architecture and GAN formulation aimed at bridging these text and image modelling advancements. Here approach is centered on translating textual concepts into vivid visual representations, effectively converting characters into pixelated images. Through rigorous experimentation, we showcase the capability of our model to produce credible images of birds and flowers from intricate textual descriptions. This work represents a significant step toward achieving the synthesis of detailed images solely from text, offering insights into the convergence of text and image modelling within the realm of artificial intelligence
Performance Evaluation of Underwater Plastic Detection Model Under Diverse Environmental Conditions
No full textProtecting the environment from ocean plastic waste requires effective detection and quantification. The number rises. This is necessary for environmental protection. Neural network-based object detectors, a recent computer vision advancement, may automate aquatic plastic detection. This computer vision innovation is new. We'll examine CenterNet HourGlass104, Faster R-CNN, and YOLOv3, three popular object detection models. This study tests these models underwater. This research examines model application.
These models were tested with a large dataset that replicated underwater conditions to determine their adaptability and conservation potential. This tested whether these models could aid conservation. This list includes plastic debris with distinct traits. Different lighting and colour conditions are included.
Our findings showed neural network object detectors' marine conservation potential. Testing showed that the CenterNet HourGlass104 was the most accurate and versatile plastic contamination detector. This applied to plastic contamination detection. Faster and more accurate plastic detection may improve cleanup efforts but harm the environment and economy. This is true even in harsh environments like underwater. These implications have major effects on real-world applications.
However, these models' limitations must be acknowledged, especially in microplastic detection and high-turbidity situations. Computer vision-based systems may be optimised and used in other ways in future research. This is because these systems' environments change.
Computer vision can detect ocean plastics, but this study highlights knowledge gaps that need further research. The study proved computer vision was possible. Due to our situation, we can help fight ocean plastic pollution. Provide detailed information about CenterNet HourGlass104, Faster R-CNN, and YOLOv3's pros and cons
Integration of Elastic Search and Kibana SIEM for Malware Detection.
No full textToday, with computers being a big part of our lives, attackers create new approaches and tools specifically aimed at these systems. Lots of papers have been written about different security methods to identify these troublemakers in company computer setups. But, now a days, significant number of these security methods are outdated, and not as effective due to these main reasons: they mainly emphasis on either network or endpoint security, lacking a structured approach; moreover, they were too simple and vulnerable. To address these shortcomings, an integrated approach that employs a combination of proactive methods and predictive threat mechanisms would result in fast detection and an immediate response using custom rules. This novel method is suitable for small and medium-sized businesses, incorporating the integration of Elastic Search and Kibana with prebuilt detection rules from Elastic. Additionally, I integrated endpoint security also created sigma rules for the purpose of detecting malware. Thus, we put together an exhaustive system for malware detection and analysis both at the network and endpoint level. In this paper, we successfully analysed a series of malware attacks using techniques from the MITRE ATT&CK matrix and was able to create custom sigma rules and alerts using the Elastic search and Kibana. I have integrated windows elastic agent collect metrics and logs from your windows machine. Then visualize that data in Kibana, create custom sigma rules by querying the logs files coming from elastic agent to create alerts for the malwares
Cognitive Stimulation Therapy (CST): Exploring Perspectives of Trained Practitioners on Barriers and Facilitators to the Implementation of CST for People Living with Dementia (Preprint)
No full textDementia is recognized as a disability under the United Nations Convention on the Rights of Persons with Disabilities (UNCRPD). People with disabilities like dementia have the right to access specialized health and social care services, including interventions that support independence and community participation. Cognitive Stimulation Therapy (CST) is an evidence-based psychosocial intervention that improves cognition, communication, confidence and quality of life for people living with dementia (PLwD), but an implementation gap means that CST is often not available. We recruited trained CST practitioners (n=62; 91.9% female) to a mixed methods study to examine facilitators and barriers to the implementation of CST in Ireland. Statistical analysis showed that 54.8% of practitioners had run CST following training; ratings of intervention efficacy predicted the likelihood of running CST groups (p=0.006); and seeing the benefits of CST first-hand predicted that practitioners would run a greater number of CST groups (p=0.01). Thematic analysis of qualitative data identified three key themes of ‘resources’, ‘awareness and education’, and ‘acceptability of CST’. Overall, the results show that while CST is acceptable and deemed highly effective, resources and staffing often impede implementation. The results are discussed in the context of prioritising the rights of people with disabilities and recommendations are made around improving access to evidence-based supports
Maximising Corporate Social Responsibility Through Purposeful Legitimacy
No full textCorporate social responsibility (CSR) has evolved into a critical component of modern business strategy, reflecting a shift towards sustainable and socially conscious practices. This study investigates the authenticity and effectiveness of CSR programmes across multiple sectors, addressing the prevalent issue of superficial and insincere initiatives in existing literature. The research aims to develop a comprehensive CSR framework that fosters genuine and impactful CSR efforts. The study employs qualitative methods, including semi-structured interviews with ten CSR professionals from diverse organizational backgrounds. This approach facilitates an in-depth exploration of the complexities and challenges associated with CSR implementation. Key themes identified include Engagement, Evaluation, Strategic Implementation, Communication and Integrity, and Dynamic Adaptation. Despite providing comprehensive insights, the study acknowledges limitations such as potential sampling bias due to the reliance on snowball sampling and the focus on organizations within the Irish business setting. These limitations suggest that findings may not be fully generalizable to different cultural or geographical contexts. The implications of this research underscore the need for a structured and authentic approach to CSR, advocating for frameworks that ensure long-term sustainability and societal impact. Future research should explore further refinement of CSR frameworks and innovative approaches to enhance the social and economic impact of CSR initiatives. This study contributes to the ongoing discourse on CSR by offering actionable insights and practical recommendations for organizations aiming to enhance their CSR efforts, thus promoting sustainable growth and societal well-being