The ISC International Journal of Information Security

Abstract Federated learning enables collaborative model training without centralized data collection, but existing frameworks rely on a central server, introducing risks of single points of failure, adversarial manipulation, and privacy leakage. To address these challenges, we propose HashLearner, a secure decentralized learning framework that utilizes the HashGraph consensus protocol for model aggregation without trusted authorities. HashLearner introduces two key innovations: (i) a consensus-driven decentralized aggregation mechanism resilient to Byzantine adversaries, and (ii) a privacy-preserving shuffling strategy that mitigates gradient reconstruction and poisoning attacks. To handle heterogeneous data distributions, the framework further employs transfer learning–based personalization. The simulation results of HashLearner, tested on benchmark Kaggle datasets, demonstrate that the platform maintains high accuracy while significantly enhancing scalability, security, and privacy. These findings indicate that HashLearner provides a practical path toward scalable, privacy-preserving, and trustworthy decentralized federated learning. 

Abstract Online social media is integral to human life, facilitating messaging, information sharing, and confidential communication while preserving privacy. Platforms like Twitter, Instagram, and Facebook exemplify this phenomenon. However, users face challenges due to network anomalies, often stemming from malicious activities such as identity theft for financial gain or harm. This paper proposes a novel method using user similarity measures and the Generative Adversarial Network (GAN) algorithm to identify anomalies (fake nodes) in user accounts in a large-scale social network while handling imbalanced data issues. Despite the problem's complexity, the method achieves an AUC rate of 80\% in classifying and detecting fake accounts. Notably, the study builds on previous research, highlighting advancements and insights into the evolving landscape of anomaly detection in online social networks. The findings of this study contribute to ongoing advancements in fake account detection, offering a hopeful solution for securing online spaces against fraudulent activities and anomaly detection in social networks.

Abstract In the realm of machine learning, Generative Adversarial Networks (GANs) have revolutionized the generation of synthetic data, closely mirroring the distribution of real datasets. This paper delves into the privacy concerns associated with GANs, particularly focusing on Membership Inference Attacks (MIAs), which aim to determine if a specific record was used in training a model. Such attacks pose significant privacy risks, especially when sensitive data is involved. To combat this, we propose a novel detector model designed to identify and thwart MIAs within GANs. Our model, which operates as an additional layer of protection for Machine Learning as a Service (MLaaS) providers, leverages outputs from both the discriminator and generator to ascertain the membership status of data samples. We introduce two variants of the detector model—supervised and unsupervised—based on the availability of information from the discriminator. The supervised detector employs labeled data for training, while the unsupervised detector uses anomaly detection techniques. Our experimental evaluation spans various GAN architectures and datasets, ensuring the robustness and generalizability of our approach. The paper also analyzes the impact of dataset size on the detector's effectiveness. By integrating our detector, MLaaS providers can enhance privacy safeguards, striking a balance between model utility and data protection.

Abstract The widespread use of web applications and running on sensitive data has made them one of the most significant targets of cyber attackers. One of the most crucial security measures that can be taken is the detection and closure of vulnerabilities on web applications before attackers. In this study, a web application vulnerability scanner was developed based on dynamic analysis and artificial intelligence, which could test web applications using GET and POST methods and had test classes for 21 different vulnerability types. The developed vulnerability scanner was tested on a web application test laboratory, which was created within the scope of this study and had 262 different web applications. A data set was created from the results of the tests performed using the developed vulnerability scanner. In this study, as a first stage, web page classification was made using the mentioned data set. The highest success rate in the page classification process was determined by 95.39% using the Random Forest Algorithm. The second operation performed using the dataset was the association analysis between vulnerabilities. The proposed model saved the 21% time than the standard scanning model. The page classification process was also used in the crawling of the web application in this study.

Abstract Sensitive methods are those that are commonly used by Android malware to perform malicious behavior. These methods may be either evasion or malicious payload methods. Although there are several approaches to handle these methods for performing effective dynamic malware analysis, but generally most of them are based on a manually created list. However, the performance shown by the selected approaches is dependent on completeness of the manually created list that is not almost a complete and up-to-date one. Missing some sensitive methods causes to degrade the overall performance and affects the effectiveness of analyzing Android malware.
In this paper, we propose a machine learning approach to predict new sensitive methods that might be used in Android malware. We use a manually collected training dataset to train two classifiers: a classifier for detecting the sensitivity nature of the Android methods, and another classifier to categorize the detected sensitive methods into predefined categories. We applied the proposed approach to a large number of methods extracted from Android API 27. The proposed approach is able to predict hundreds of sensitive methods with accuracy of 90.5% for the first classifier and 87.4% for the second classifier. To evaluate the proposed approach, we built a new list of the detected sensitive methods and used it in a number of tools to perform dynamic malware analysis. The proposed model found various sensitive methods that were not considered before by any other tools. Hence, the effectiveness of these tools in performing dynamic analysis are increased.

Abstract A Chatbot is a smart software that responds to natural language input and attempts to hold a conversation in a way that simulates humans. Chatbots have the potential to save any individual’s time, hassle, and tedium by automating mundane tasks. The idea of this research is that to investigate how to help the user efficiently interact with the robot receptionist through an Intelligent Assistant dialogue. Chatbots are an effective way to improve services with their 24 /7 uptime, and their cost efficiency, and their multi-user quality. Despite the chatbots reduce human errors and give more answers that are accurate. Successful implementation of a chatbot requires correct analysis of the user’s query by the bot and ensures the correct response that should be given to the user. This research develops a chatbot for the Airports, which provides the visitors to the SWE chatbot Relevant information about the department. Throughout our extensive search since the very begin- ning of our project, we have been through multiple re- sources and endured a strenuous vetting process.

Abstract The increasing volatility in pricing and growing potential for profit in digital currency have made predicting the price of cryptocurrency a very attractive research topic. Several studies have already been conducted using various machine-learning models to predict crypto currency prices. This study presented in this paper applied a classic Autoregressive Integrated Moving Average(ARIMA) model to predict the prices of the three major cryptocurrencies âAT Bitcoin, XRP and Ethereum âAT using daily, weekly and monthly time series. The results demonstrated that ARIMA outperforms most other methods in predicting cryptocurrency prices on a daily time series basis in terms of mean absolute error (MAE), mean squared error (MSE) and root mean squared error(RMSE).

Abstract Recent improvements in web standards and technologies enable the attackers to hide and obfuscate infectious codes with new methods and thus escaping the security filters. In this paper, we study the application of machine learning techniques in detecting malicious web pages. In order to detect malicious web pages, we propose and analyze a novel set of features including HTML, JavaScript (jQuery library) and XSS attacks. The proposed features are evaluated on a data set that is gathered by a crawler from malicious web domains, IP and address black lists. For the purpose of evaluation, we use a number of machine learning algorithms. Experimental results show that using the proposed set of features, the C4.5-Tree algorithm offers the best performance with 97.61% accuracy, and F1-measure has 96.75% accuracy. We also rank the quality of the features. Experimental results suggest that nine of the proposed features are among the twenty best discriminative features.