This research presents a comprehensive comparison of hybrid models combining K-Nearest Neighbours (KNN) with Support Vector Machines (SVM) and Long Short-Term Memory (LSTM) networks for stock price prediction. Using historical data from major technology stocks (AAPL, GOOGL, MSFT) spanning 2020-2023, we implemented and evaluated KNN-SVM and KNN-LSTM architectures. Our experiments demonstrate varied performance across different stocks, with the KNN-SVM hybrid achieving optimal results for GOOGL (RMSE: 2.11, R2: 0.968) while showing moderate performance for AAPL (RMSE: 13.56) and MSFT (RMSE: 25.39). This variation in performance provides important insights into the relationship between stock characteristics and model efficacy. Through extensive experimentation and architectural optimization, we demonstrate that while simpler hybrid models can achieve superior performance in specific cases, their effectiveness varies significantly across different market conditions and stock characteristics. The study reveals important insights into the trade-offs between model complexity, computational efficiency, and prediction accuracy in financial time series forecasting.

Key management is a critical aspect of securing Wireless Sensor Networks (WSNs), ensuring confidentiality, integrity, and authentication of the communicated data. This paper presents a comprehensive study on key management techniques in WSNs, focusing on both static and dynamic methods. Static key management involves pre-distributing keys to sensor nodes before deployment, offering simplicity and reduced communication overhead but facing challenges in scalability and resilience against node capture attacks. Dynamic key management, on the other hand, entails periodic updates and re-establishment of keys, enhancing security through adaptability and resilience but at the cost of increased communication and computational overhead. We compare and contrast various static methods, such as random key pre-distribution and pairwise key establishment, with dynamic methods like re-keying protocols and key refreshment techniques. Our analysis highlights the strengths and weaknesses of each approach in terms of security, efficiency, and practicality for different WSN applications. By providing a detailed exploration of key management schemes, this paper aims to guide the design and implementation of secure and efficient key management protocols tailored to the unique constraints and requirements of WSNs.

This paper introduces a Federated Orchestration Framework that integrates P4-ShieldNet for core network security with SAFE-CAST for edge security, addressing the complex challenges in modern telecommunication networks. Our approach combines programmable data planes with distributed intelligence, providing a comprehensive security solution from core to edge. Simulations demonstrate significant improvements over traditional methods and standalone implementations, achieving 97% threat detection accuracy, 80% faster response times, and support for up to 50,000 devices. The framework incorporates cross-domain security coordination, federated learning, and intent-based policies, paving the way for adaptive, self-defending networks suitable for next-generation 5G and 6G infrastructures.

With the advent of data driven era, new challenges have emerged that can hinder their efficient and reliable utilization. At the center of this era is data. However, public, and unrestricted access to data may pose privacy and security threats. Federated learning methods have proposed to address this problem. While the main concern in federated learning has been preserving privacy of data and maintaining the efficiency of centralized learning methods, the structure and distribution of data has not been studied sufficiently. In this paper, we present a federated machine learning approach capable of accommodating imbalanced datasets. Our method dynamically adjusts the learned parameters to prevent biased classifications. Our empirical results indicates that the proposed method managed to improve accuracy by 22%.

As the number of academic publications increase, preparing a literature review becomes more challenging. This paper introduces an automated literature review support system to ease the literature review process for academia with reference graphs, abstract and full document summaries, paper clusters by keywords, abstracts, and abstract summaries combined. The output of the proposed system may ease exploring the state-of-the-art research.

The development of e-commerce has created new shopping trends of customers. In online shopping environments, product reviews play a critical role in the choice of customers. Online reviews are additionally valuable for the manufacturers and the vendors by providing easily accessible feedback to them. In this study, a text analysis method is proposed to find the defective features of the products by detecting features with negative opinion tendency in the clustered customer reviews. The output of the proposed model, the extracted defects, may provide a strong source of guidance both for consumers in purchase decisions and for producers in product improvement.

Social media platforms are places where people post their feelings and thoughts about a topic. The institutions, organizations, individuals, or companies that are the subject of these ideas are affected by these posts. As discussed in different studies, companies in stock exchange markets are affected by the posts made on these social media platforms. At the same time, individuals who are aware of this fact, namely speculators and influencers, may make profit by manipulating the truth. In this study, possible speculators or influencers using the Twitter social media platform are investigated. As the target companies, Google, Amazon, Apple, Tesla, and Microsoft were chosen, which are among the largest companies on the NASDAQ stock exchange market. In the study, asentiment analysis using the Loughran and McDonald sentiment analysis dictionary was utilized. The sentiment analysis results were used to model different machine learning algorithms. With the models, individuals who had too many positive or negative effects as possible speculators or influencers were identified. The study was performed for 5 years of data. The results indicates that (1) without noise reduction, it is not possible to establish a correlation on individual tweets and their effects on the stock market; (2) it is not possible to establish a correlation between the number of tweets and the volume of companies; (3) the effect of threshold on the accuracy, which has been done and proven in different studies, has also been proven in this study; (4) RBF Kernel SVM method gives better result than other machine learning methods.

We present the project SpEnD, a complete SPARQL endpoint discovery and analysis portal. In a previous study, the SPARQL endpoint discovery and analysis steps of the SpEnD system were explained in detail. In the SpEnD portal, the SPARQL endpoints are extracted from the web by using web crawling techniques, monitored and analyzed by live querying the endpoints systematically. After many sustainability improvements in the SpEnD project, the SpEnD system is now online as a portal. SpEnD portal currently serves 1487 SPARQL endpoints, out of which 911 endpoints are uniquely found by SpEnD only when compared to the other existing SPARQL endpoint repositories. In this portal, the analytic results and the content information are shared for every SPARQL endpoint. The endpoints stored in the repository are monitored and updated continuously.

Trolls in social media are `malicious' users trying to propagate an opinion or distort the general perceptions. Identifying trolls in social media is a task of interest for many big data applications since data cannot be analyzed effectively without eliminating such users from the crowd. In this paper, we present a solution for troll detection and also the results of measuring terror awareness among social media users. We used Twitter platform only, and applied several machine learning techniques and big data methodologies. For machine learning we used k-Nearest Neighbour (kNN), Naive Bayes, and C4.5 decision tree algorithms. Hadoop/Mahout and Hadoop/Hive platforms were used for big data processing. Our tests show that C4.5 has a better performance on troll detection.

Providers (SP) are wishing to increase their Return of Investment (ROI) by utilizing the data assets generated by tracking subscriber behaviors. This results in the ability of applying personalized policies, monitoring and controlling the service traffic to subscribers and gaining more revenues through the usage of subscriber data with ad networks. In this paper, a framework is developed to monitor and analyze the Internet access of the subscribers of a regional SP in order to categorize the subscribers into an interest category from The Interactive Advertising Bureau (IAB) categories. The study employs the categorization engine to build category vectors for all subscribers. The simulation results show that once a subscriber has been classified into a category the click rate for the same subscriber group can be improved by correlating the interests of the subscribers with the advertisements.

In this paper, we propose a novel hybrid topology and interest based clustering to organize the overlay network in order to reduce startup latency and service interruption probability. Our method uses a clustering technique to organize the peers according to their interests and locality information. The proposed technique is compared with the current unstructured P2P architectures in terms of average hit time, hit ratio for searched content, and the maximum rate of information (in bits per second) that can be transmitted over P2P network caused by protocol and overlay- level connectivity. The simulation results show that the proposed system outperforms the current unstructured P2P architectures.