Context The problem of story point estimation still remains a challenge in agile software development. Statistical methods and conventional machine learning models are not effective in understanding the nuances of the tasks. Even though LLMs excel in analyzing and producing texts that are linguistically complex, their capacity for performing real semantic reasoning is still controversial. As far as effort estimation is concerned, their forecasts will continue to be inaccurate if they lack project-related context. Objective This study aims to investigate the effectiveness of retrieval-based augmented LLMs in the context of estimating story points with increased accuracy compared to traditional statistical approaches and LLMs. Methods In this paper, a Retrieval-Augmented Large Language Model framework that combines semantic similarity search with the text-based inference capabilities of a large language model is presented. To validate the performance of the framework, the TAWOS dataset consisting of 23,313 tasks across 17 projects is used. In the evaluation process, a stratified splitting approach based on the project level is used. To assess the performance, Pred(25), MAE, RMSE, MMRE, and MdMRE metrics are used. To test the significance, the Wilcoxon signed-rank test is applied. Results The proposed model achieves a Pred(25) value of 37.18%, outperforming the world average (34.53%), project median (27.79%), project average (25.67%), and vanilla LLM (27.96%). Furthermore, there is an improvement in MAE and RMSE, implying increased robustness and reduced occurrences of outlier estimation errors. Paired statistical tests show significantly lower values for estimation error (p < 0.001), but effect-size calculations point to negligible and small practical effects, depending on the benchmark. At the project level, performance is project-specific, which implies that the utility of using retrieval-enhanced effort estimation depends on the availability of similar historical efforts. Conclusion Research results indicate that grounding story points for estimation using contextual information via retrieval-based reasoning facilitates effort estimation. The observed improvement is associated with the incorporation of retrieval-based project context. The findings suggest that story point estimation can be more effectively supported when treated as a retrieval-grounded contextual reasoning task rather than as a purely statistical prediction problem.
Task allocation is an important aspect of the multi-UAV system, as it has a direct impact on the efficiency, scalability, and reliability of the mission. However, the growing need for UAV swarms to perform missions in complex and heterogeneous environments, such as disaster response, maritime rescue, and edge services, has created new challenges to the traditional task allocation methods, especially with respect to scalability and adaptability. This paper provides a comprehensive analysis of the problem of task allocation for the multi-UAV system, including the problem formulation, system models, and the proposed solution methodologies. In particular, we examine the optimization-based, coalition-based, auction-based, security-based, cost-based, and learning-based methods for the problem of task allocation for the multi-UAV system. Based on the analysis, we reveal the trade-offs between the different methods and identify some of the important issues, such as the trade-offs between the different aspects of the problem, such as robustness and adaptability, security and complexity, scalability and communication, with the aim of providing guidelines for the development of new frameworks for the problem of task allocation for the multi-UAV system.
In the complex mission environment of Unmanned Aerial Vehicle (UAV) swarms, it is imperative to develop efficient task allocation methods that can be feasibly executed in the presence of heterogeneous UAV operation constraints. This study has developed a competitive hyper-heuristic framework for UAV task allocation that can concurrently execute multiple meta-heuristic optimization methods and select the most balanced task allocation strategy through a normalized multi-objective cost function. The task allocation problem has been modeled as a constrained optimization problem that can handle heterogeneous UAV operation constraints, different types of tasks, time constraints, and resource constraints. It can also optimize travel distance, mission time, energy consumption, priority satisfaction, and workload distribution. The framework uses the Genetic Algorithm, Particle Swarm Optimization, Grey Wolf Optimizer, Simulated Annealing, and Greedy Local Search methods within a fixed computational budget to eliminate algorithm selection biases. It uses a penalty normalization technique to prevent the dominance of constraint violation in the objective function. The experimental results for a mission scenario with six heterogeneous UAVs and twenty spatially distributed tasks show that the UAV task allocation problem does not necessarily minimize individual efficiency metrics for the most feasible task allocation strategy. For example, the Grey Wolf Optimizer has the shortest travel distance of 562.5 km and the least energy consumption. However, the Particle Swarm Optimization algorithm has the least total cost of 0.75 due to the near-optimal performance of the algorithm in solving the problem while minimizing constraint violation.
In the context of the accuracy, dependability, and security of application programming interfaces (APIs), the importance of API testing is undeniable. Nevertheless, the effort required to comprehensively test APIs is significant, especially when creating valid API test cases. This study seeks to develop a framework that automates the generation of API test cases while ensuring the structural validity and functional correctness of the API test cases. The study aims to reduce the effort required by human testers to generate API test cases. The study proposes a framework that utilizes a large language model (LLM) to generate API test cases based on the concept of structural similarity. The proposed framework utilizes the LLM to generate API test cases while ensuring the functional correctness of the API test cases. To validate the structural similarity between the API test cases generated by the proposed framework, the study introduces the Jaccard similarity-based API test case structural validation mechanism. The study utilizes a multi-domain API ecosystem that contains more than 40 endpoints to validate the effectiveness of the proposed API test case generation framework. The study found that the proposed API test case generation framework achieved a high success rate of 84% in generating valid API test cases for different API domains. The study found that the proposed API test case generation framework using the LLM achieved a high mean Jaccard similarity coefficient of 0.827, indicating the high level of structural alignment between the API test cases generated by the LLM and the API specifications. Therefore, the study concludes that the proposed API test case generation framework using the LLM is effective in automating the generation of API test cases while ensuring the accuracy, dependability, and security of the API test cases.
The advancement of Large Language Model (LLMs) has opened new possibilities for automating software development, particularly in generating test codes. This study examines LLM capabilities for autonomously generating Robot Framework test codes using a Retrieval-Augmented Generation (RAG) model. By leveraging a repository with stored contextual information, the model enhances the accuracy and relevance of generated test scripts while reducing human intervention. To assess the generated tests, we employ CodeBLEU to evaluate syntactic correctness, Jaccard Similarity to measure structural alignment, and pass/fail statistics from CI/CD pipelines to determine real-world applicability. Our experimental results demonstrate that integrating RAG with LLMs significantly improves both test quality and reliability. Contextual information retrieval enhances semantic cohesion and functional robustness, while a multi-metric evaluation framework ensures a comprehensive assessment. This study highlights the transformational role of LLMs in software testing automation, emphasizing their potential to streamline continuous integration (CI) practices. By enhancing test code generation accuracy and minimizing manual effort, the proposed approach marks a significant step toward fully automated and contextually aware testing within the Robot Framework.
The increasing complexity of Internet of Drones (IoD) platforms demands more accessible ways for users to interact with unmanned aerial vehicle (UAV) data systems. Traditional methods requiring technical API knowledge create barriers for non-specialist users in dynamic operational environments. To address this challenge, we propose a retrieval-augmented generation (RAG) architecture that enables natural language querying over UAV telemetry, mission, and detection data. Our approach builds a semantic retrieval index from structured application programming interface (API) documentation and uses lightweight large language models to map user queries into executable API calls validated against platform schemas. This design minimizes fine-tuning needs, adapts to evolving APIs, and ensures schema conformity for operational safety. Evaluations conducted on a curated IoD dataset show 91.3% endpoint accuracy, 87.6% parameter match rate, and 95.2% schema conformity, confirming the system’s robustness and scalability. The results demonstrate that combining retrieval-augmented semantic grounding with structured validation bridges the gap between human intent and complex UAV data access, improving usability while maintaining a practical level of operational reliability.
The efficacy of Network Intrusion Detection Systems (NIDS) is often challenged by the high dimensionality of modern network traffic data, which can degrade classification accuracy and increase computational overhead. To address this, this paper proposes a novel ensemble feature selection model that integrates three distinct meta-heuristic algorithms: Genetic Algorithm (GA), Grey Wolf Optimizer (GWO), and Particle Swarm Optimization (PSO). The proposed model not only leverages the individual exploratory strengths of each optimizer but also amalgamates their findings through an advanced, fitness-based weighted voting mechanism. This approach mitigates the risk of converging to a local optimum and aims to produce a more robust and generalizable feature subset. The system is designed as a fully automated, end-to-end pipeline that dynamically identifies the best-performing feature set and proceeds with further model tuning, ensuring reproducibility and efficiency. The model's performance was validated on the X-IIoTID dataset. Our empirical results confirm the model's superior performance: the Ensemble model consistently achieved the highest classification accuracy compared to individual GA, GWO, and PSO methods. Furthermore, the evaluation duration of the ensemble model was notably the lowest. This efficiency allows for quicker anomaly detection and response, crucial for real-time NIDS applications. By effectively reducing feature space dimensionality while enhancing or maintaining classification accuracy, our automated pipeline provides a robust and reproducible solution for developing more effective and computationally efficient NIDS.
The growing use of unmanned aerial vehicles (UAVs) in diverse fields such as disaster recovery, rural regions, and smart cities necessitates effective dynamic drone network establishment techniques. Conventional optimization techniques like genetic algorithms (GAs) and particle swarm optimization (PSO) are weak when it comes to real-time adjustment to the environment and multi-objective constraints. This paper proposes a hybrid optimization framework combining genetic algorithms and reinforcement learning (RL) to improve the deployment of drone networks. We integrate Q-learning into the GA mutation process to allow drones to adaptively adjust locations in real time under coverage, connectivity, and energy constraints. In the scenario of large-scale simulations for wildfire tracking, disaster response, and urban monitoring tasks, the hybrid approach performs better than GA and PSO. The greatest enhancements are 6.7% greater coverage, 7.5% less average link distance, and faster convergence to optimal deployment. The proposed framework allows drones to establish strong and stable networks that are dynamic in nature and adapt to dynamic mission demands with efficient real-time coordination. This research has important applications in autonomous UAV systems for mission-critical applications where adaptability and robustness are essential.
The Internet of Drones (IoD) is transforming autonomous aerial missions in applications like disaster response, agriculture, and infrastructure monitoring. Traditional Unmanned Aerial Vehicle (UAV) decision-making systems are typically challenged by real-time responsiveness, safety, and interpretability. In this study, we propose a new hybrid decision-making system that fuses Large Language Models (LLMs) with Retrieval-Augmented Generation (RAG) for improved UAV autonomy. By leveraging real-time sensor readings and past mission logs, our system allows drones to make contextual decisions with greater reliability and interpretability. Our suggested methodology is realized in an IoD platform, where LLMs produce adaptive commands by learning context-dependent information from a structured knowledge graph. Experimental tests, performed in both simulation and real-world settings, show that our framework greatly enhances decision accuracy. The LLM-RAG system realized 92
Reliable analysis of UAV telemetry data is critical for mission safety, especially as drones are increasingly deployed in complex and high-risk environments. These data streams often include anomalies arising from sensor faults, environmental disruptions, or cyber-physical attacks, making robust anomaly detection essential. This study introduces an unsupervised anomaly detection framework designed specifically for high-frequency UAV telemetry. It combines domain-driven feature engineering with an AutoML-based optimization pipeline that enables automated model selection and hyperparameter tuning. The framework integrates four unsupervised algorithms—Local Outlier Factor, Isolation Forest, One-Class SVM, and Elliptic Envelope—ensuring adaptability to the dynamic nature of UAV operations. Evaluated on a real-world dataset of 127,000 samples from 48 UAV missions, the system uses expert-labeled anomaly segments solely for validation to preserve the integrity of unsupervised learning. Among all methods, Local Outlier Factor yielded the best results with 0.920 accuracy, 0.880 precision, 0.850 recall, and 0.860 F1-score. Scalable and low-latency, the proposed solution is well-suited for real-time deployment. By bridging theoretical advances with operational needs, this work contributes to safer and more resilient aerial robotic systems.
Unmanned Aerial Vehicles (UAVs) are now valuable tools in as varied fields as logistics, surveillance, and environmental monitoring. With the growth of drone-based operations, energy efficiency optimization in mission planning is increasingly a key concern-especially for battery-powered short-endurance UAVs. Drone selection is one issue deserving thoughtful examination of both mission parameters, payload demands, and platform-dependent limitations, and the most appropriate drone likely to be successful at a given mission. To overcome this difficulty, we suggest a regression-based energyaware drone selection technique. The system compares three supervised learning algorithms-Decision Tree, Random Forest, and Linear Regression-each of which is optimized for hyperparameters. The model with the highest accuracy, selected using $\mathbf{R}^{2}$ and RMSE measures, is used to forecast energy consumption based on attributes like drone type, payload, and mission distance. The system was trained and validated using a database of $\mathbf{1, 0 0 0}$ mission records. Results show that mission range and payload weight have the largest impact on energy consumption. The model always assigns drones to minimize energy consumption while meeting operational requirements. Fixed-wing drones are most appropriate for long-range missions, while multirotors are most suitable in short-range, high-accuracy missions. This study is the basis of large-scale intelligent UAV fleet management systems that are optimized for energy consumption.
Honeypots have long served as decoy systems in cybersecurity, luring attackers to gather intelligence on their techniques. However, traditional honeypots face challenges like scalability, data overload, and susceptibility to detection. AI-powered honeypots, utilizing machine learning and reinforcement learning, address these limitations by dynamically adapting to evolving threats, improving detection rates, and engaging attackers for longer periods. This paper explores the evolution of honeypots from static systems to AI-driven defenses, highlighting their role in mitigating complex cyberattacks such as IoT botnet-driven DDoS attacks. We compare the strengths and weaknesses of AI-based honeypots with traditional models, examine their performance metrics, and discuss ongoing challenges related to data quality, evasion techniques, and ethical considerations.
As cyberattacks get more sophisticated, security systems must learn to detect and deceive them. DecoyPot, a honeypot Web Application Programming Interface (API) that generates legitimate API responses, is introduced in this paper. DecoyPot's command extractor module carefully analyzes API requests to create prompt-response pairs that improve a Retrieval-Augmented Generation based (RAG) large language model (LLM). DecoyPot can instantly adjust its answers to mimic API activity in a contextually correct and convincing manner to attackers. To assess system efficacy, we used a two-phase similarity analysis. Initial queries were matched with prompt-response pairs to ensure contextually suitable responses. Second, similarity measures were used to compare generated responses to reference responses, producing an average score of 0.9780. The high score shows that the system can create API-like responses, boosting its utility. DecoyPot engaged opponents and learned their Tactics, Techniques and Procedures (TTPs). The study shows that honeypot cybersecurity effectiveness must be improved by merging AI-driven response creation with enhanced deception technologies. DecoyPot effectively adapts to incoming queries and generates API-like responses, delivering actionable cyber threat intelligence and enhancing proactive defense strategies.
The efficient deployment of drones to establish an effective communication network is a challenging problem in a variety of use cases, from disaster management to rural coverage. In this study, we present a hybrid optimization strategy using Ant Colony Optimization (ACO) and Deep Reinforcement Learning (DRL) for optimizing drone placement and mobility in a target geographic area. The proposed method leverages ACO's global search capability coupled with DRL's adaptive learning capability for optimizing network coverage and guaranteeing optimal connectivity among drones and a central hub. The hybrid technique is contrasted with a solo ACO approach, with the former exhibiting superior performance regarding coverage, connectivity, and deployment efficiency.
The increasing sophistication of cyberattacks makes Intrusion Detection Systems (IDSs) essential, yet the high dimensionality of modern network traffic hinders accuracy and efficiency. We conduct a comparative study of multi-objective feature selection for IDS using four bio-inspired metaheuristics—Grey Wolf Optimizer (GWO), Genetic Algorithm (GA), Particle Swarm Optimization (PSO), and Ant Colony Optimization (ACO)—on the X-IIoTID dataset. GA achieved the highest accuracy (99.60%) with the lowest FPR (0.39%) using 34 features. GWO offered the best accuracy–subset balance, reaching 99.50% accuracy with 22 features (65.08% reduction) within 0.10 percentage points of GA while using ~35% fewer features. PSO delivered competitive performance with 99.58% accuracy, 32 features (49.21% reduction), FPR 0.40%, and FNR 0.44%. ACO was the fastest (total training time 3001 s) and produced the smallest subset (7 features; 88.89% reduction), at an accuracy of 97.65% (FPR 2.30%, FNR 2.40%). These results delineate clear trade-off regions of high accuracy (GA/PSO/GWO), balanced (GWO), and efficiency-oriented (ACO) and underscore that algorithm choice should align with deployment constraints (e.g., edge vs. enterprise vs. cloud). We selected this quartet because it spans distinct search paradigms (hierarchical hunting, evolutionary recombination, social swarming, pheromone-guided foraging) commonly used in IDS feature selection, aiming for a representative, reproducible comparison rather than exhaustiveness; extending to additional bio-inspired and hybrid methods is left for future work.
The Internet of Drones (IoD) integrates autonomous aerial platforms with security, logistics, agriculture, and disaster relief. Decision-making in IoD suffers in real-time adaptability, platform interoperability, and scalability. Conventional decision frameworks with heuristic algorithms and narrow Artificial Intelligence (AI) falter in complex environments. To mitigate these, in this study, an augmented decision model is proposed, combining large language models (LLMs) and retrieval-augmented generation (RAG) for enhancing IoD intelligence. Centralized intelligence is achieved by processing environment factors, mission logs, and telemetry, with real-time adaptability. Efficient retrieval of contextual information through RAG is merged with LLMs for timely, correct decision-making. Contextualized decision-making vastly improves adaptability in uncertain environments for a drone network. With LLMs and RAG, the model introduces a scalable, adaptable IoD operations solution. It enables the development of autonomous aerial platforms in industries, with future work in computational efficiency, ethics, and extending operational environments. In-depth analysis with the collection of drone telemetry logs and operational factors was conducted. Decision accuracy, response time, and contextual relevance were measured to gauge system effectiveness. The model’s performance increased remarkably, with a BLEU of 0.82 and a cosine similarity of 0.87, proving its effectiveness for operational commands. Decision latency averaged 120 milliseconds, proving its suitability for real-time IoD use cases.
One of the most critical steps toward autonomous capabilities, where natural language instructions can be successfully converted into executable API calls, is integrating Large Language Models (LLMs) into the ecosystem of the Internet of Drones (IoD). This study introduces an end-to-end LLM-based framework for enhancing real-time drone operation and problem handling in intent recognition, parameter extraction, and ambiguity resolution. It has resorted to a spectrum of methodologies in the form of Retrieval-Augmented Generation (RAG) and customized fine-tuning specific to each domain, towards accurate command interpretation and successful API generation. This paper evaluates the performance of the model on a carefully designed dataset containing 1,500 commands for the different scenarios of IoD with an average BLEU score of 89.6 and a cosine similarity of 0.94. Optimization techniques, such as parallel processing and better query handling, reduced the latency of this system by 15%, with an average query processing time of 0.9 seconds. This work gives considerable importance to the scalability and flexibility of the system, which is quite crucial for applications in domains like disaster response, precision agriculture, and surveillance. Proposed LLM-based framework thus tries to bridge gap between human intent and drone execution for intuitive, reliable, and efficient IoD deployments.
Unmanned Aerial Vehicles (UAVs) are used in different fields ranging from recreational vehicles to agriculture, environmental monitoring, infrastructure inspection, disaster management, security, surveillance and logistics. This paper provides an overview of UAV applications and highlights the importance of object detection in improving drone autonomy. Object detection facilitates tasks such as precision agriculture, disaster response, environmental protection, infrastructure inspection and autonomous navigation. The paper reviews recent advances in UAV-specific object detection algorithms and methodologies and highlights challenges such as varying altitudes, motion blur, real-time processing constraints and limited computational resources. Solutions and adaptations are discussed to overcome these challenges, including lightweight neural network architectures, transfer learning, data augmentation, edge computing, multi-sensor fusion, attention mechanisms and adaptive algorithms. The study highlights the importance of integrating object detection systems into UAVs to improve their use in various sectors and contribute to improvements in surveillance missions.
Industrial Internet of Things (IIoT) is a rapidly growing field, where interconnected devices and systems are used to improve operational efficiency and productivity. However, the extensive connectivity and data exchange in the IIoT environment make it vulnerable to cyberattacks. Intrusion detection systems (IDS) are used to monitor IIoT networks and identify potential security breaches. Feature selection is an essential step in the IDS process, as it can reduce computational complexity and improve the accuracy of the system. In this research paper, we propose a hybrid feature selection approach for intrusion detection in the IIoT environment using Shapley values and a genetic algorithm-based automated preprocessing technique which has three automated steps including imputation, scaling and feature selection. Shapley values are used to evaluate the importance of features, while the genetic algorithm-based automated preprocessing technique optimizes feature selection. We evaluate the proposed approach on a publicly available dataset and compare its performance with existing state-of-the-art methods. The experimental results demonstrate that the proposed approach outperforms existing methods, achieving high accuracy, precision, recall, and F1-score. The proposed approach has the potential to enhance the performance of IDS in the IIoT environment and improve the overall security of critical industrial systems.
Through interconnecting machines, sensors, and devices, the Industrial Internet of Things (IIoT) has revolutionized industrial processes by allowing real-time data analysis and decision-making. The implementation of cloud, edge, and fog computing models has followed IIoT development, bringing unique security and privacy challenges in their wake. In this all-encompassing report, the potential risks, complications, and apprehensions of these models are analyzed with respect to IIoT integration. Examining cloud, edge, and fog computing models distinguishes and assesses their benefits when applied to IIoT. The paper analyzes the strengths offered by each model and addresses the risks associated with security and privacy, such as data infringement, malware, and denial-ofservice attacks. IIoT systems, in particular, pose significant challenges in providing protection. Cloud, edge, and fog computing models provide ample benefits but IIoT systems utilizing these models must have strong security and privacy policies in place. This paper highlights specific technical and organizational strategies to tackle potential challenges and risks. By effectively implementing these strategies, organizations can tap into the advantages of these models while ensuring the safety of their IIoT systems. Informed decisions can be made, and IIoT deployments can be safeguarded with the aid of the valuable resource given in this paper, which offers comprehensive analysis and recommendations. For those in the field of IIoT security, including practitioners and researchers, this paper is of great value.