1,720,998 research outputs found
Towards Explainable Deep Learning Models for Fault Prediction based on IoT Sensor Data
No full textThis thesis addresses a pressing issue in the realm of IoT-based fault prediction using sensor data, focusing on the crucial yet challenging aspect of explainability within deep learning models. While deep learning has showcased remarkable advancements in fault prediction, its inherent black-box nature poses obstacles in understanding the rationale behind its predictions. This lack of transparency impedes the practical implementation and adoption of these models in critical decision-making scenarios. The thesis comprises a comprehensive investigation encapsulated within five published papers spanning over a decade, from 2011 to 2023. These papers collectively contribute to the domain of Explainable Artificial Intelligence (XAI), delving into various approaches aimed at shedding light on the inner workings of complex deep learning models. The earlier papers serve as building blocks, laying the groundwork for fundamental concepts explored and expanded upon in subsequent submissions. Each paper makes distinct contributions to the field of AI. These contributions include the introduction of a novel evolutionary algorithm, applying Fuzzy Cognitive Maps for failure and fault modelling, proposing an evolutionary algorithm for training Fuzzy Cognitive Maps, developing an explainable deep learning model for fault prediction, and utilizing insights derived from preceding research to explaining the inner processes of deep learning models. Through a meticulous analysis of these publications, this thesis effectively addresses the fundamental research questions posed. It offers insights into overcoming the challenges associated with the opacity of deep learning models, paving the way for more transparent and interpretable AI models, particularly in the domain of fault prediction using IoT sensor data
Proceedings of the fifth UK/BCS symposium on knowledge discovery and data mining
No full textThis is the proceedings of a one day symposium on Knowledge Discovery and Data Mining held at the Salford Lowry in 2009. The topics covered included some of the most important and exciting issues in the field. There were presentations on fundamental research topics such as how data mining is changing the very nature of scientific methods, the challenges of time series data mining, use of social network analysis for classification of messages, knowledge discovery from case data, and development of a unifying framework for feature selection methods. There were also presentations describing the lessons learned from real world case studies in detecting financial crime, profiling electricity usage, image processing, credit scoring, and predicting internet shopping pattern
Knowledge based improvement:simulation and AI for improving unplanned maintenance operations
No full textUsing Knowledge Discovery and Data Mining Methods to Detect Price Manipulations in Financial Markets
No full textUsing Knowledge Discovery and Data Mining Methods to Detect Price Manipulations in Financial Markets
No full textUsing Knowledge Discovery and Data Mining Methods to Detect Price Manipulations in Financial Markets
No full textAutomatic Classification, Detection and Segmentation of Breast Arterial Calcification on Digital Mammography Images Using Deep Learning
No full textCardiovascular disease (CVD) is the leading cause of premature death in the United Kingdom with one type, coronary artery disease, killing more than twice as many women as breast cancer. Conventional CVD risk factors have been shown to have less accuracy for females who are considered low-risk. Recently, researchers have noted that breast arterial calcification (BAC), which is regularly observed as an incidental finding on mammograms, could be used to risk-stratify women for CVD.In 2023, almost 2 million women attended breast screening clinics in England. Automatic BAC detection on mammograms could provide vital additional cardiovascular information, without the need for further invasive tests or radiation exposure, and could direct patients to relevant clinical pathways or therapies.As a first step in automating the BAC grading process, I developed deep learning models for BAC classification, object detection and segmentation using an anonymised dataset which was annotated for the presence and location of BAC under the guidance of two consultant radiologists. Data augmentation was used in both the classification and object detection networks, increasing the training data size.My modified ResNet22 network showed promise in classifying the presence or absence of BAC at image level, attaining a test accuracy of 80%, indicating that this method could be used as a simple flag for this purpose. I also used this network for feature extraction in Faster R-CNN and YOLO BAC object detection models. Despite improving on a recent similar study, these latter networks performed poorly with very low average precision scores at several thresholds. As an improvement, this study developed a DeepLabv3+-based BAC segmentation network which doubled the IoU obtained by another study using a similar model and achieved a BFScore of over 70% specifically for BAC.Based on the findings of this research, a two-step pipeline is recommended with our classifier triaging mammographic images for BAC and our segmentation model providing an indication of the extent of its presence. This could provide the basis for further research in order to realise the potential of concurrent, automatic BAC grading for women undergoing mammographic imaging
Enhancing Cybersecurity: Machine Learning and Natural Language Processing for Arabic Phishing Email Detection
No full textPhishing is a significant threat to the modern world, causing considerable financial losses. Although electronic mail has shown to be a valuable asset around the world in terms of facilitating communication for all parties involved, whether huge corporations or individuals communicating in their everyday lives, it has also brought with it its own set of issues. Scammers take advantage of such issues by sending out bogus emails to susceptible persons in order to acquire access to their personal information. Phishing email detection is considered an important research field, and the research community has tried hard to address this problem in various common languages like English. There are some other important languages, such as Arabic, which have not been given much attention when it comes to phishing detection. Arabic is the native language of more than 300 million people and is ranked as the fifth most extensively used language throughout the world. In terms of content-based phishing email detection, there has been relatively little research on Arabic language phishing emails. This study presents an English-Arabic Phishing Detection (EAPD) model developed on the word level (Term Frequency-Inverse Document Frequency (TF-IDF), Document-Term Matrix (DTM), and FastText embedding) and the character-level convolutional neural network (CharEmbedding) to decrease this gap. It will be one of the first studies to explore the extent to which machine learning (ML) and natural language processing (NLP) methods can be used to develop models for detecting English/Arabic phishing attacks. An English-Arabic parallel phishing email corpus was developed using the English and Arabic text provided by the leading security and privacy analytics anti-phishing shared task (IWSPA-AP 2018). To evaluate the effectiveness of the EAPD model, a collection of balanced 1258 emails in Arabic and English, featuring equal ratios of legitimate and phishing emails, was used. The experiments indicate that when using the Multilayer Perceptron (MLP) classifier combined with TF-IDF, the EAPD achieved an accuracy of 95.3% on Arabic datasets. The English text, on the other hand, reached a 95.7% accuracy when paired with the Support Vector Machine (SVM) classifier and TF-IDF. Salloum's list, a new set of Arabic stop words, was introduced and found that while traditional ML classifiers remained largely unaffected, deep learning (DL) models with FastText embedding, especially LSTM, showed a significant 14% variance following the integration of this extended list. Overall, this study presents a promising approach for detecting phishing emails in both English and Arabic, with high accuracy and efficiency
Going Beyond Counting First Authors in Author Co-citation Analysis
No full textThe present study examines one of the fundamental aspects of author co-citation analysis (ACA) - the way co-citation
counts are defined. Co-citation counting provides the data on which all subsequent statistical analyses and mappings
are based, and we compare ACA results based on two different types of co-citation counting - the traditional type that
only counts the first one among a cited work's authors on the one hand and a non-traditional type that takes into
account the first 5 authors of a cited work on the other hand. Results indicate that the picture produced through this non-traditional author co-citation counting contains more coherent author groups and is therefore considerably clearer. However, this picture represents fewer specialties in the research field being studied than that produced through the traditional first-author co-citation counting when the same number of top-ranked authors is selected and analyzed. Reasons for these effects are discussed
- …
