The issue of online fake news is a major concern because many people rely on online platforms for updated information and news. With the massive amount of information on these platforms, fake news identification has become a daunting task. Users often rely on the reactions and comments of others in their social circle on these platforms to judge the credibility of news stories. Many researchers are actively working to detect online fake news by analyzing post titles and headlines. However, there is a considerable research gap as people's comments are currently not being utilized for fake news detection on platforms such as Facebook and Instagram. Integrating people's comments on such news into the fake news detection process is crucial in effectively detecting fake news. To address this challenge, we gather and organize two separate datasets FNID (Fake News Instagram Dataset) and FNFD (Fake News Facebook Dataset) containing news titles and comments. These datasets are specifically created to aid in the detection of online fake news. Subsequently, we propose an innovative hybrid model for news titles and users' comments - DFN-SCNC (Detection of fake news based on social context and news content) - that combines BERT and Bi-GRU models. Our model, which combines BERT for tokenizing and extracting contextual vectors with Bi-GRU for analyzing post content and social interactions (comments), outperforms various state-of-the-art techniques achieving an F1-score of 97% and 91 % on FNID and FNFD datasets respectively.
Electricity demand will rise by $\mathbf{5 0 \%}$ by the year 2035. This will put pressure on the electricity suppliers to shift towards renewable energy sources (RESs) such as wind (WT) and solar (PV). The shift from traditional energy sources is driven by the exploitation of conventional energy and the impact of global warming. This research presents an optimization model integrating RESs like WT and PV with a battery storage system. The main goal of this study is to schedule system sources and optimize the sizing of the system components for smooth system operation and cost-effectiveness. With the existing JAYA and GWO techniques, a new hybrid optimizer setup called the JayaGrey Wolf Optimizer (JGWO) is adapted for unit sizing. The combined objectives of minimizing net annual cost (TAC) and loss of supply power probability (LPSP) are achieved for a standalone (SA) microgrid (MG) environment. The proposed research demonstrates that using a hybrid multi-objective optimization approach, the JGWO results in substantial cost savings and improved grid reliability compared to JAYA and GWO. Hence, this makes JGWO a suitable approach for this research problem.
The study aims to cope with the drawbacks incurred in the conventional User Requirement Analysis (URA) through a new framework that integrates Sentiment Analysis, clustering, and Support Vector Machine (SVM) classifiers. This approach's goal is to optimize URA by categorizing user requirements based on the sentiment of specific target websites. It also demonstrates that these approaches outperform conventional methods in terms of accuracy, recall rates, and speed, enabling more effective consideration of clients' needs in problem-solving processes. In general, this research will explore the potential of machine learning to optimise and enhance the URA process, while also highlighting potential issues that could complicate the requirement gathering process.
Radio Frequency Identification (RFID) is one of the most promising authentication technologies in pervasive computing. Besides identification, the RFID also offers mutual authentication to IoTs (Internet of Things) terminal devices and manages the access control of IoT networks. To decrease the overall cost of the IoT sensors (RFID tags), Ultralightweight Mutual Authentication Protocols (UMAPs) have been proposed. These UMAPs incorporate simple bitwise logical operators (e.g. XOR, AND, OR, Modulo 2 addition, etc.) and extremely low-cost non-triangular primitives in their designs to ensure confi-dentiality, authentication, and integrity. In this paper, we have performed a cryptanalysis of the recently proposed UMAP: Ultralightweight RFID Authentication Protocol (URAP). We have exploited the weak structure of URAP design and identified a desynchronization attack. The proposed attack requires only two (2) sessions to desynchronize the IoT sensor from its network fully. We have also discussed other design flaws of the protocol structure which can lead towards several full disclosure attacks.
The fair management of water shortage and its global distribution posture are substantial issues, with the water situation in Palestine being made worse by the IsraeliPalestinian conflict. The study highlights how critical it is to conduct in-depth research to forecast water demand patterns while accounting for resource limitations and the difficulties posed by conflict. The primary objective is to analyze water usage in Palestine using predictive analytics, with attention to different industries and regions. The research investigates historical data, examining significant aspects such as Year, Period, Memory Number, Consumption, Area No., Area Name, and Type. The scope contains an extensive dataset analysis that recognizes patterns and trends in water usage using cuttingedge statistical and deep learning algorithms. This work is important since it is the first to apply a Hybrid model that combines ARIMA and Multi-Layer Perceptron (MLP). Timeseries data is used in this innovative strategy to get precise predictions. The Hybrid model is applied to predict water usage in Palestine across different geographical areas and customer groups. The proposed hybrid model produced an MAE, MSE, and RMSE value of 0.0663, 0.0063, and 0.0798 respectively.
Traditional energy meters, namely electronic and electromechanical meters have been used longer to measure electricity consumption. These old technology meters lack realtime monitoring and automatic control mechanisms, making the system inefficient and energy expensive. This paper proposes an IoT-based control panel designed to connect to the output of the traditional old energy meters, evolving them into smart devices capable of real-time data monitoring, visualization, and automated switching control. The proposed system incorporates smart sensors, micro-controller, and communication module to measure and monitor the voltage and current, calculate power and energy, and switch between energy meters based on the consumption limits or thresholds, which trigger the increase in price of energy consumed. This advancement grasps the step tariffs to maintain the energy consumption within lower-cost slabs, reducing overall electricity costs. The system is also capable of protecting household appliances during over and undervoltages. The device also features, IoT functionality, enabling remote data monitoring through mobile and web apps. The implementation of this smart panel demonstrates an efficient and cost-effective solution for household energy management, offering a significant reduction in power tariffs and enhancing energy efficiency.
Organizations delivering digital services and products to their customers are facing a number of different challenges due to the evolving landscape of cyber threats. To overcome these challenges, an increasing number of these firms are investing in secure software development. This research study explores the impact of the following software security best practices on software product security and quality. For this purpose, first, an online survey is conducted in the Pakistani software industry to determine the most commonly used software security best practices. Based on the information obtained from this survey (34 responses from 32 different organizations), it is found that some of the most widely used software security practices are Authentication, Authorization, and Accountability; Secure Deployment Environment; Application Programming Interface (API) and Component Security Testing; and Error Handling Mechanism and Logging. Next, an industrial experiment is designed and conducted to determine the impact of these practices on real-world projects. Keeping practical considerations in mind, 5 of the top most widely-used security best practices are shortlisted for this experiment and 3 different software development companies are engaged. Before starting the experiment, training on these shortlisted security best practices is also provided to the representatives of these companies. Software quality and software size data of two consecutive sprints – sprint n (without security best practices) and sprint n+1 (with security best practices) – is collected from each of these 3 companies. Metrics such as weighted defect density and weighted security defect density are calculated for both sprints. A comparative analysis reveals a significant improvement in software security and overall software quality in sprint n+1 vis á vis sprint n.
Cyber-Physical Systems (CPS) integrate networking, computing, and physical processes, forming the backbone of critical industries such as healthcare, energy, and transportation. The increasing complexity and interconnection of CPS have led to significant compliance and security challenges. This research introduces a novel framework that leverages Machine Learning (ML) techniques to enhance access control, authorization, and accountability within CPS environments. By combining these techniques with traditional access control methods, the framework addresses the unique demands of CPS, including scalability and adaptability to dynamic conditions. A key innovation lies in applying ensemble methods like Random Forest, AdaBoost, and Gradient Boosting, which outperform individual models by mitigating overfitting and improving generalizability. The framework also incorporates sophisticated feature engineering and regularization strategies tailored to CPS, ensuring robust and efficient security solutions. Through rigorous data preprocessing, relationship analysis, and model validation, this study demonstrates how machine learning can significantly advance the security posture of CPS, offering a scalable and effective approach tailored to the specific needs of these critical systems.
Tracking human actions, especially in the field of medicine, has encouraged interest in computer vision, specifically for detecting gait anomalies to aid in rehabilitating patients with irregular patterns. In this paper, the proposed system employs vision-based and inertial-based modalities that enable the recording of human actions and encourage human-machine collaboration. The inertial sensors were preprocessed using a Butterworth filter, as its magnitude response is often monotonic and maximally flat in the passband, providing smoothness. Multiple key features, such as Mel Frequency Cepstral Coefficients (MFCC) and auto-regression coefficients have been extracted. On the other hand, human detection has been performed for vision-based data, and main features such as angles and velocities have been calculated. The final results have been fused using multimodal survey fusion. The fused data was then forwarded for optimization using yeo-johnson power optimizer. Artificial Neural Networks (ANN) have been trained to perform classification. Evaluation using the confusion matrix on the Heriot-Watt University of Sao Paulo (HWU-USP) dataset proficient of identifying nine complex actions showed an accuracy rate of 83.33%.
Shift frequency jamming (SFJ) is a sophisticated electronic countermeasure (ECM) technique, to disrupt the operation of contemporary linear frequency modulated (LFM) radars. This deception jamming method manipulates the intercepted radar signal to introduce multiple false targets, and retransmits it back to the target radar. In this paper, an effective electronic counter-countermeasure (ECCM) technique is proposed to neutralize the effects of SFJ. The proposed antijamming technique is based on incorporating variations in the chirp rate of LFM waveform. The effectiveness of the proposed technique is justified by mathematical formulation, and extensive simulations in different scenarios. The proposed anti-jamming methodology efficiently extracts the true target from multiple SFJ false targets. Finally, numerical results and analysis are carried out to demonstrate the authenticity of the proposed ECCM framework.
Aspect-based sentiment analysis (ABSA) seeks to extract fine-grained sentiment information by locating sentiments connected to particular elements in a text. Traditional sentiment analysis techniques often fail to address the intricacies of aspect-based sentiments due to their inability to capture complex con-textual dependencies. The suggested approach makes use of long short-term memory networks (LSTMs) to describe sequential dependencies and convolutional neural networks (CNNs) for effective feature extraction. Simultaneously, the attention mech-anism improves interpretability and performance by focusing on pertinent portions of the input text. The CNN component extracts local features from word embeddings, which the LSTM then processes to capture long-term dependencies. By giving important words a larger weight, the attention mechanism further refines the result, ensuring that the most informative parts of the text are emphasized during sentiment and aspect prediction. Extensive experiments, including Automobile, Movie, and hotel reviews, use benchmark datasets to demonstrate the efficacy of our model. The collaborative framework significantly outper-forms baseline models, improving F1 scores, recall, accuracy, and precision.
Addressing vehicle identification and classification within dynamic traffic monitoring systems requires advanced methodologies that can overcome the limitations of traditional approaches in data integration and computational efficiency. This study introduces a comprehensive framework for vehicle classification in aerial image sequences, utilizing a sequential pipeline of georeferencing, preprocessing, segmentation, YOLOv8-based detection, and feature extraction through SIFT and FAST algorithms. Classification is performed using a CNN-BiLSTM hybrid model, which has been rigorously validated to ensure high accuracy and robustness. This comprehensive method significantly enhances vehicle identification and classification efficacy, rendering it very relevant for practical traffic management and surveillance applications. Experimental validation of the Vehicle Aerial Imagery from a Drone (VAID) dataset demonstrates the framework's superior performance, attaining an accuracy of 0.968% and exceeding current methodologies in the aerial classification of vehicles.
The ability to process text has been effectively increased with the development of AI and Natural Language Processing (NLP) techniques. There is a growing need for efficient legal assistance due to the rise in court proceedings, particularly those related to Intellectual Property Rights (IPR). This research aims to develop a novel legal search assistance system to predict legal judgments and extract relevant data from Trademark cases as well as Trademark Ordinance based on user's input query. For judgment forecasting of legal scenarios, XGBoost, SVM and Random Forest (RF) were used, with the mean cross validation score as 67%, 59% and 56%. The use of pre trained BERT model in the designed system further enhances the efficiency of data retrieval. In terms of cases and ordinance data extraction, Mean Average Precisions (MAP) of PAK-LEGAL-BERT and legal-bert-base ranges between 67 % to 71 %, while after fine tuning these models, MAP values increases from 85% to 95% effectively. A user-friendly GUI makes the system accessible to all, which helps in paper drafting of legal cases, irrespective of the legal expertise.
Automated clinical dialogue summarization can help make health professional workflows more efficient. With the advent of large language models, machine learning can be used to provide accurate and efficient summarization tools. Generative Pre-Trained Transformers (GPT) have shown huge promise in this area. While larger GPT models, such as GPT-4, have been used, these models pose their own problems in terms of precision and expense. Fine-tuning smaller models can lead to more accurate results with less computational expense. In this paper, we fine-tune a GPT-3.5 model to summarize clinical dialogue. We use both default hyperparameters along with manual hyperparameters for comparison purposes. We also compare our default model to past work using ROUGE-1, ROUGE-2, ROUGE-L, and BERTScores. We find our model outperforms GPT-4 across all measures. As our fine-tuning process is based on the smaller GPT-3.5 model, we show that fine-tuning leads to more accurate and less expensive results. Informal human observation also reveals our notes to be of acceptable quality.
In the context of the growing demand for wireless communication technologies such as WiFi and Bluetooth, the development of a robust and efficient RF power amplifier is of paramount importance. In this context, effective impedance matching is essential to minimize signal distortion and maximize power efficiency. This study emphasizes the critical role of effective impedance matching in contemporary RF power amplifier design, essential for sustaining the increasing demand for wireless communication technologies like WiFi and Bluetooth. This work combines stub matching methods with multiple segments of Multi Quarter-Wave Transformers (QWT), presenting a strategic approach to achieve optimal outcomes. Our study provides a detailed exploration of both stub matching and the architecture of Multi QWT sections within the amplifier, offering valuable insights to address this fundamental challenge in contemporary RF power amplifier design.
Wireless and mobile network complexity is increasing at an unheard-of rate. Traditional network control and management techniques, which rely on analytical models and simulations, are becoming more and more unworkable as a result of this rapid expansion. When we plan for the data influx from cutting-edge applications like augmented reality, especially in the realm of software-defined wireless local area networks (SD-WLANs), this difficulty becomes even more obvious. We support a forward-looking strategy in which future SD-WLANs adopt an AI-native perspective to meet this growing complexity. The Ai-WiFi platform, based on AI, is described in this article as a means of enabling autonomous SD-WLAN management. This paper proposes an innovative and comprehensive model for software-defined wireless local area networking (SD-WLAN) that integrates the usage of AI and packet length aggregation. Our idea is in line with the most recent developments in in-network AI and, by emphasizing the improvement of frame sizes in SD-WLANs, we demonstrate the viability of Ai-WiFi in a selected use case. We also consider the future development, difficulties, and prospective possibilities for AI-assisted network management in the context of wireless networking.
Software Effort Estimation (SEE) is an important component of software project planning. Many projects fail due to inaccurate software estimates. Over-commitment is considered one of the factors that contribute towards inaccurate estimates. This study presents a mechanism to adjust software effort estimates for over-commitment. At step 1, the study identifies - through a pilot study and literature review- the factors that affect the SEE. A list of 43 factors affecting SEE is prepared from the existing literature and the pilot study. Then an industrial survey (Survey I) is conducted to know the extent to which each of these factors can affect the effort estimates. Results of the survey show that the most crucial factors affecting the SEE are over promising and over-commitment. At step 2, the study presents an approach to adjust existing effort estimates for over-commitment. These adjustments are proposed based on the Survey II results. The Survey II indicates the total percentage of effort estimates that are affected by over commitments as well as ten factors of over commitments. At step 3, the study validates the proposed adjustment approach through an experimental survey (Survey III); around 77.08% of the respondents considered the adjustments good. During the validation step, we collected information regarding the estimated and the actual effort spent. The results of the evaluation show that on average, the effort estimates were off by 20.24%. After applying the proposed adjustments this percentage was reduced to 17.99% on average. This reduction indicates that the proposed adjustments have improved the accuracy of the effort estimates by 11.12% approximately. The fact that 17.99% estimates are still inaccurate indicates there is still a need to have more adjustment factors for SEE.
In wireless communication, localization plays a key role in different applications such as asset tracking, navigation and emergency response. This paper explores a deep learning framework with feature reduction techniques to improve localization accuracy. We employ a dataset of received wireless signals at different base stations and perform feature reduction using three methods: Principal Component Analysis (PCA), Random Forest Feature Importance (RF) and Recursive Feature Elimination (RFE). Next using the processed features, a deep neural network (DNN) is utilized to predict the localization coordinates. By lowering computational complexity and preventing overfitting our findings show that feature reduction can greatly enhance the DNN's performance comparable with no feature reduction. Further, the REF was found to be the most effective feature-reduction technique for training of DNN model in terms of localization accuracy. Extensive experiments demonstrate that feature reduction and deep learning can be combined to achieve accurate localization in wireless communication systems.
Biogas Power Plants are a relevant source of renewable energy, producing biogas from anaerobic fermentation of almost any biomass and making the gas available, e.g. by providing gas for heating and cooking, or by providing electricity from combined heat and power (CHP) plants. Nevertheless, the control of the fermentation process and the resulting biogas production is still a largely unsolved problem, leading to a lack of flexibility in both the processable input biomass and the resulting biogas production. Most available control technology for the fermentation process is either costly, too complex and not sufficiently robust for operation in a biogas power plant, or not delivering practical value or better yield. This is partly due to the difficulties with developing mathematical plant models and controllers. Here comes AI into play which can lead to usable predictions of non-linear, complex and insufficiently understood processes. This paper presents the ongoing research on a low-cost in-situ measurement system for online operation in the substrate of a biogas fermenter which predicts the process parameters and gas production with AI methods. The solution is targeted for retrofitting the large number (> 9,000) of existing agricultural biogas power plants in Germany. In addition, the approach is expected to support the operation of smaller biogas power plants in rural regions by non-experts, leading to a more sustainable development of the energy infrastructure and the efficiency of agriculture in such regions, both from ecological and economical viewpoint.
Gliomas are the most common and severe type of brain tumor that causes higher mortality in adults worldwide. Accurate segmentation of brain tumors is crucial for treatment planning and patient survival. Given the heterogenetic nature of brain tumors, like indistinct boundaries and variability in size, shape, and location, this is a real challenge in medical imaging. In this research, we use an effective dual feature extraction strategy that integrates features from the U-Net encoder with the vision transformer encoder. UN et is effective at extracting local features, while Transformer excels at global feature extraction due to its higher receptive field. The final segmentation map is generated using the decoder part of the UN et. The proposed model generates satisfactory results. On BRATS 2020, dice similarity coefficients of 89.57%, 79.97%, and 82.44% for whole tumor, enhanced tumor, and tumor core, segmentation, respectively. Ablation results on BRATS 2021: Dice similarity coefficients of 92.5%, 84.75%, and 89.2% for the same tasks. With an average IOU/Jaccard coefficient of 80.80%, an average dice of 87.08%, a Hausdorff distance (95th percentile) of 4.68 mm, and an average surface distance of 0.90 mm. The model complexity is lower than other state-of-the-art methods which makes this more practical in this domain.
