Internet of Things (IoT) systems are beneficial to our daily lives and have become increasingly important. A complete IoT system includes devices, sensors, networks, software, and other essential components necessary for operation and interconnection. However, devices and sensors of this nature often have low resource requirements and multiple security vulnerabilities from manufacturers. Moreover, edge network areas of IoT systems exhibit several security weaknesses. Consequently, unauthorized hijacking of sensors or denial-of-service attacks on edge network areas can have severe consequences for the system’s operation. In this study, we propose a model that combines machine learning algorithms and principal component analysis techniques to train and predict Distributed Denial of Service (DDoS) attacks. Principal component analysis techniques were applied to reduce data dimensionality. We used accuracy, precision, recall, and F1-Score as the evaluation metrics. We explain the True Positive, False Positive, True Negative, and False Negative measures as basic parts of the above evaluation metrics. Unlike previous studies, we used the Training Time to evaluate the training time of each model. We employed two datasets, CICIDS 2017 and CSE-CIC-IDS 2018, to evaluate our proposed model. In general, the proposed models exhibited the best performance and improved training time.

The growth of information technology has resulted in a massive escalation of data and the demand for data exploration, particularly in the machine learning sector. However, machine learning raises concerns about data privacy because algorithms require large amounts of data to learn and make accurate predictions. Such data often contain personal information about individuals, and there is a risk that this information could be accessed or misused by unauthorized parties. It is crucial for organizations that use machine learning to prioritize personal data protection and ensure that appropriate safeguards are in place to prevent privacy breaches. Federated learning (FL) and blockchain technology are two increasingly popular approaches to distributed computing. Federated learning is a distributed machine-learning approach that trains machine-learning models on decentralized datasets without centralizing the data. Federated learning offers several benefits, including improved data privacy. Ensuring the benefit of clients in federated learning is vital for the success of this distributed machine learning approach, especially when combined with blockchain technology, as it offers a secure and transparent way to store and verify data. In this study, we propose a combination of federated learning and blockchain as a solution to some of the challenges faced by both approaches. By leveraging the decentralized nature of federated learning and the security and transparency of blockchain, our approach tends to overcome issues such as data privacy and trustworthiness of results. The evaluation results demonstrated that the proposed approach has many potential applications in various domains.

Smart contracts are autonomous programs stored on blockchain networks that self-execute agreed terms in a transparent and accurate manner. Within cross-chain platforms, smart contracts facilitate interaction and exchange of data between diverse blockchains. However, the presence of vulnerabilities in smart contracts renders them susceptible to exploitation, jeopardizing security. Considerable research has focused on identifying and detecting such vulnerabilities, though existing approaches have yet to achieve comprehensive coverage. This paper presents ChainSniper, a sidechain-based framework integrating machine learning to automatically appraise vulnerabilities in cross-chain smart contracts. A comprehensive dataset, denoted "CrossChainSentinel", was compiled comprising 300 manually labeled code snippets. This dataset was leveraged to train machine learning models discerning vulnerable versus secure smart contracts. Experimental findings demonstrate the viability of machine learning methodologies for enhancing smart contract auditing within decentralized applications spanning multiple networks. Notable detection precision was achieved, substantiating ChainSniper’s potential to strengthen security analysis through an automated and expansive evaluation of smart contract code.

Although online auctions have gained popularity as a marketplace format, they encounter challenges stemming from bidders withdrawing high bids without penalty, and wasting organizer resources due to redundant auction rounds. This behavior, known as overbidding, undermines efficiency and remains a longstanding issue in traditional auction designs. It’s worth noting that one of the common reasons for bid withdrawals is exceeding their financial capacity. Participants bid beyond their budgetary constraints, not necessarily with the goal of obtaining the asset but for alternative motives. In an effort to address these overbidding inefficiencies, this paper presents Zk-Auction, a decentralized blockchain-based cross-chain auction system. Zk-Auction leverages the privacy-preserving capabilities of zero-knowledge proofs to cryptographically verify bidder funds for each bid, deterring inflation beyond actual holdings. A novel sidechain architecture simulates interoperability between independent "Banking" and "Auction" blockchains, enabling bid authentication without revealing identities or balances. By tackling the core challenges of existing centralized auction platforms, Zk-Auction adopts a distributed off-chain matching and on-chain settlement approach to maintain security and interoperability across heterogeneous distributed ledgers. The integration of Zero-Knowledge Proofs in bid validation ensures a privacy-preserving manner. Additionally, a proof-of-concept prototype demonstrates feasibility at scale. Evaluations of the system show enhanced participant accountability, thereby incentivizing sincere bidding behavior in contrast to traditional designs that are susceptible to wasteful overbidding.