Unmanned Aerial Vehicles have become indispensable assets across various sectors, leveraging their mobility and data collection capabilities. However, privacy and security concerns have fueled interest in Federated Learning as a potential solution. Federated learning (FL), being decentralized and collaborative, holds promise in addressing the privacy risks inherent in centralized data processing while enhancing model performance. In this review, we explore the privacy and security implications of FL in UAV ecosystems. We highlight FL’s potential to mitigate privacy risks by aggregating model updates locally, minimizing data transmission needs. In addition, we examine security challenges and evaluate protective mechanisms. Through a systematic review of the literature, we identified gaps and proposed future research directions, with the aim of improving the security and privacy of FL in UAV applications. Our findings highlight advanced strategies, such as secure aggregation, adversarial defenses, and lightweight cryptographic techniques, to mitigate privacy and security threats in Federated Learning-based UAV. They also underscore the need for real-world validations and regulatory frameworks to ensure resilient and ethically governed deployments in UAV networks.

Continuing development of network technologies and the escalation of cyber threats have heightened the the requirement for IDS or Intrusion Detection System solutions. There are two issues that may remain a challenge for IDS techniques developed within the scope of traditional approaches: the accuracy of detection and applicability of measures. the face of evolving threats. This paper focus on comparing ten machine learning algorithms (Artificial Neural Network (ANN), K-Nearest Neighbors (K-NN), Naive Bayes (NB), Random Forest (RF), Decision Tree (DT), Logistic Regression (LR), Support Vector Machine (SVM), Gradient Boosting (GB), AdaBoost (AB), and XGBoost) using the CICIDS2017 dataset, which represents a wide array of network intrusions in traditional network environments. The main purpose of this research is to measure the efficiency of these algorithms from different learning attacks in order to develop a real-time intrusion detection system. The performance of each model is further evaluated by examining some metrics values. The results show that XGBoost, followed by other ensemble based models, that include Random Forest and Gradient Boosting, provide almost 100\% detection rates, making them some of the best models to tackle large and skewed data sets. These algorithms were able to achieve high accuracy across the board and also in identifying the minority attack classes that are so relevant in a good IDS. On the other hand, there were under-performed algorithms such as Naive Bayes and AdaBoost to discover specific types of attacks suggesting their infeasibility when used for intricate intrusion detection problems without further enhancement. Traditional algorithms such as SVM and Logistic Regression showed good overall accuracy but faced challenges with minority classes, highlighting the need for advanced techniques to improve their performance. This analysis underscores the strengths and limitations of different machine learning algorithms in detecting sophisticated cyber threats within traditional network architectures. The findings provide valuable insights for selecting optimal models in designing robust and scalable intrusion detection systems for current and next-generation networks. Future work should focus on addressing class imbalance and enhancing model generalization to improve detection rates for underrepresented attack types, thereby strengthening the overall security posture against emerging cyber threats.