Authors - Mohd. Zuhaib Ahmed, Akash Priya, Deepti Chopra, Pankaj Kumar Abstract - Effective landing and take-off (LTO) decision-making in mil itary aviation is critically dependent on airfield serviceability and pre vailing weather conditions. A fundamental challenge is the absence of structured expert pilot decision logs, as such data are operationally sen sitive and access-restricted. This work presents a replicable methodolog ical framework for developing machine learning-based decision support systems in domains where operational data are scarce or classified. The pipeline encompasses synthetic data forged using correlated Monte Carlo sampling, constrained by location-specific geographic, seasonal, and ter rain parameters across ten Indian Air Force (IAF) stations, yielding ap proximately 60,000 simulated operational scenarios. The dataset is gen erated within domain-constrained operational bounds to ensure physi cal plausibility. A rule-based expert classification system assigns opera tional status as Green (Safe), Orange (Caution), or Red (Unsafe); four ML algorithms are subsequently evaluated: Logistic Regression, Naïve Bayes, Support Vector Machines, and Decision Trees. The Decision Tree achieves the highest performance, with an accuracy of 0.983, an F1 score of 0.983, and a ROC-AUC of 0.984. The proposed framework supports two deployment pathways: the rule engine as a deterministic automa tion tool for standard clearances, and the ML model as the inference core of a real-time Human-in-Loop (HIL) expert system requiring opera tor authorisation at every decision. As expert pilot decision logs become available, the system may be progressively elevated to a fully adaptive expert system.
Authors - Ritesh Kumar Verma, Preethiya T Abstract - Contemporary customer support systems require processing a massive number of user queries with low latency and high semantic relevance. Rule-based systems fail to capture context, while fully LLM-based systems are computation ally expensive and suffer from high latency. This paper introduces an adaptive AI-assisted customer support automation system using an optimized Retrieval Augmented Generation (RAG) model. The proposed system combines Azure OpenAI embeddings, FAISS-based vector search, selective Cross-Encoder re ranking, and a Learning-to-Rank (LambdaMART) model for adaptive score fu sion. Unlike vanilla RAG models, the proposed system adaptively re-ranks only the top-k retrieved candidates, trading off ranking precision and latency. Experi ments were carried out on a 1,30,000-sample e-commerce customer support da taset with query-response pairs annotated with intent labels. Compared to rule based retrieval, embedding+FAISS, and vanilla RAG models, the proposed hybrid system showed improved top-1 retrieval precision with a concurrent reduc tion in end-to-end latency from 0.414s to 0.365s (≈11.8% relative improvement). The LambdaMART model adaptively learned weights from FAISS and Cross Encoder scores, improving ranking robustness and eliminating misranked top re sponses. The system was implemented on Azure Machine Learning with a cloud scale pipeline and interactive Streamlit web interface, showcasing the cost-effec tive inference capabilities of the proposed system via selective re-ranking.
Authors - Abdelrahman El Antably, Ali Hamdi, Ammar Mohamed Abstract - Large Language Models (LLMs) frequently generate plausi ble but incorrect information, known as hallucinations. Detecting these errors at a fine-grained level is crucial, especially for morphologically rich languages like Arabic with limited resources. We introduce BAL ANCE:Bi-perspective Analysis for LLM Accuracy via coNsensus ChEck ing, a novel dual-judge framework for token-level hallucination detection in Arabic LLM outputs. Our six-module pipeline features context filtra tion, argument decomposition, and distinct strict and lenient LLM-based judges. A consensus coordinator then synthesizes their verdicts, and a span annotator precisely localizes errors. Evaluated on the Arabic sub set of the SemEval-2025 MuSHROOM benchmark, BALANCE achieved an Intersection over Union (IoU) score of 72.87%. This significantly outperforms the task’s winning system by approximately 8.76% rela tive improvement and consistently surpasses zero-shot baselines across various LLMs by up to 39.80 percentage points.
Authors - Duy Pham, Tung-Duong Le-Duc, Anh-Tai Pham-Nguyen, Trung Nguyen Mai, Long Nguyen, Dien Dinh Abstract - Multimodal knowledge graphs improve structured knowledge representation and tasks such as cross-graph entity alignment. However, most benchmarks focus on resource-rich languages and assume dense relational structures and balanced attributes. Low-resource languages like Vietnamese pose additional challenges, including structural sparsity, attribute asymmetry, and modality noise. To address this gap, we in troduce DBWiki-VN15K, the first large-scale Vietnamese multimodal knowledge graph dataset for entity alignment. Built from Wikidata and DBpedia, it contains 15,000 aligned entity pairs with relational triples, lo calized numerical attributes, and visual modalities. The dataset provides both word-segmented and unsegmented text to support different linguis tic processing approaches. Experiments with state-of-the-art multimodal entity alignment models reveal that structure-guided multimodal fusion and dynamic modality weighting are more robust to sparse and noisy features. Additionally, unsegmented subword tokenization better han dles cross-graph translation inconsistencies than strict Vietnamese word segmentation. DBWiki-VN15K offers a realistic benchmark for studying multilingual and multimodal knowledge fusion. Our dataset is available at: https://github.com/Tim50c/DBWiki-VN15K.
Authors - Ritesh Jawarkar, Reena Satpute, Sudhir Agarmore Abstract - Because sleep problems can influence the health of a person and his/her quality of life, such diagnosis and treatment relies on specific classification. Even though single deep learning and machine learning models have shown their potential, they are limited by overfitting and bias in the model. In order to solve these issues, the current research proposes the expansion of the ensemble learning-based sleep disorder classification through the inclusion of machine learning model predictions. A voting classifier enhances the optimization base classifier outputs in terms of robustness and classification accuracy. According to Sleep Health and Lifestyle Dataset, the ensemble method has 97.3 percent accuracy with individual models. The interface is designed as a Flask-based web interface that allows user authentication to increase user interaction and usage of the system on a real-time basis. Suggested extension ensures the reliable, accurate and easy-to-use automated sleep problem diagnosis.
Authors - Aman Kumar, Mary Subaja christo Abstract - Content Delivery Networks (CDNs) play an essential role in enhancing the content delivery speed by caching frequently requested data in edge servers distributed across geographical regions. Traditional CDNs utilize rule-based policy and machine learning approaches for optimizing the cache. Machine learning is performed centrally, and the cache optimization is performed using the traffic logs collected by the central server. Although the use of central learning approaches is beneficial, it poses certain limitations, including data privacy and high communication cost. The central learning approach aggregates raw data, which poses data privacy issues. This paper proposes an architecture for secure federated learning, which is utilized for cache hit prediction in CDNs. The proposed architecture is evaluated using a synthetic dataset containing 1,30,548 records, and the features include temporal and network features. The proposed architecture is compared with the traditional central learning approach, and the results reveal that the secure federated learning model achieves an accuracy of 70.15%, which is comparable to the central learning approach. The proposed architecture is found to reduce data privacy exposure by 30%.
Authors - Bambang Marsudi Salim, Hudan Studiawan, Baskoro Adi Pratomo Abstract - Digital forensic investigations face a growing threat from sophisticated log tampering, in which adversaries delete or modify computer event logs to conceal evidence of criminal activity. This paper presents an empirical comparison of A Search and Iterative Deepening A* (IDA*) for reconstructing falsified computer event logs, extending the previous bipartite graph framework. Three log artefacts were constructed from the public forensic timeline dataset: an original computer log, a trusted ISP log, and a deliberately falsified log containing 15 strategically deleted events. To address timestamp heterogeneity arising from different system and ISP browser log parsers, a window-based matching strategy is introduced. Experiments conducted across maximal consecutive event sequences (MCES) demonstrate that IDA* consistently explores fewer nodes than A*. Anomaly detection identified 60.7% of browser events as uncorroborated by ISP records, achieving 60.0% recall on the 15 deliberately deleted events.
Authors - Akshay Ladha, Supraja P Abstract - Twitter social media platforms have become the primary means of communication for customer support, requiring rapid, accurate, and scalable response solutions. Conventional customer support mechanisms are primarily manual and inefficient in handling large volumes of real-time interactions. This paper presents an AI-Assisted Twitter Support System that combines deep learning with distributed streaming engines to automate real-time customer interactions. The system design utilizes Apache Kafka for tweet streaming, Apache Spark Streaming for distributed processing, and Long Short-Term Memory (LSTM) networks for sentiment analysis and multi-class complaint classification. A confidence-aware decision-making module is used to ensure that automated responses are produced only when the prediction confidence level exceeds certain thresholds, thus avoiding potential miscommunications. The system was trained and tested on the Kaggle Airline Sentiment dataset (1,46,400 tweets) with three sentiment classes and eight complaint categories. The sentiment analysis model attained an accuracy of 85.2% (F1-score of 0.846), and the complaint classification model attained an accuracy of 80.5% (F1-score of 0.792). The complete pipeline maintained an average latency of 2.9 seconds with a maximum processing rate of 2500 tweets per minute.
Authors - Pravitha N R, Sreelakshmi S R, Valsalachandran K, Savithri S Abstract - The rapid expansion of digital services has significantly increased the collection and processing of personal data through online platforms such as e-commerce systems, social media applications, and digital payment services. To regulate the use of personal information, governments worldwide have introduced data protection regulations such as the General Data Protection Regulation (GDPR), the Digital Personal Data Protection Act (DPDPA), and the California Consumer Privacy Act (CCPA). Organizations publish privacy policies to inform users about their data practices; however, these policies are often lengthy, complex, and difficult for users to understand. Consequently, users frequently accept privacy policies without fully reviewing how their personal data is collected, processed, and shared. Recent research has explored automated approaches for privacy policy analysis using artificial intelligence techniques, including machine learning, natural language processing, and large language models. Retrieval-Augmented Generation (RAG) has further enhanced compliance evaluation by linking policy statements with relevant regulatory clauses. Despite these advancements, challenges remain, such as the lack of standardised datasets, limited explainability of AI decisions, dependence on prompt design, and insufficient validation with regulatory experts. This paper discusses future research directions in AI-driven privacy policy compliance analysis and highlights emerging opportunities for improving regulatory compliance assessment, user privacy protection, and transparent privacy governance in digital ecosystems.
Authors - Ayushi Raj, Malathy C Abstract - The rapid growth of sensitive data requires backup systems that are both storage-efficient and risk-aware. Traditional backup approaches rely on static policies that ignore temporal changes, data sensitivity, and redundancy, leading to inefficient storage use and higher risk exposure. This work proposes a risk-adaptive backup optimization framework integrating temporal modelling, sensitivity-aware deduplication, and online learning. The system reconstructs data evolution using intrinsic timestamps and quantifies data criticality through continuous sensitivity scoring. A unified risk model combines sensitivity, change intensity, and exposure over time to determine backup urgency. An online rein forcement learning agent dynamically optimizes backup decisions based on evolving data patterns. The framework applies secure, sensitivity-based dedupli cation to reduce redundancy while preserving privacy. Operating in a read-only, metadata-driven manner, it ensures compliance with strict data governance re quirements. By decoupling decision logic from storage, the system supports hy brid cloud environments. Experimental results show reduced storage costs and controlled risk, demonstrating its effectiveness for scalable, intelligent data pro tection.
Authors - Nimisha K, Sridharan G, Kathiresh kumar K, Lohit S, Shyam Ganesh K Abstract - The rapid growth of sensitive data requires backup systems that are both storage-efficient and risk-aware. Traditional backup approaches rely on static policies that ignore temporal changes, data sensitivity, and redundancy, leading to inefficient storage use and higher risk exposure. This work proposes a risk-adaptive backup optimization framework integrating temporal modelling, sensitivity-aware deduplication, and online learning. The system reconstructs data evolution using intrinsic timestamps and quantifies data criticality through continuous sensitivity scoring. A unified risk model combines sensitivity, change intensity, and exposure over time to determine backup urgency. An online rein forcement learning agent dynamically optimizes backup decisions based on evolving data patterns. The framework applies secure, sensitivity-based dedupli cation to reduce redundancy while preserving privacy. Operating in a read-only, metadata-driven manner, it ensures compliance with strict data governance re quirements. By decoupling decision logic from storage, the system supports hy brid cloud environments. Experimental results show reduced storage costs and controlled risk, demonstrating its effectiveness for scalable, intelligent data pro tection.
Authors - Anirudh P, Nimisha K, Princy P Abstract - As technology advances, circuit complexity increases, integrated cir cuits become more prone to defects during manufacturing and operation. Conse quently, in order to ensure reliable operation, effective testing and stability eval uation of memory cells are essential. Static random-access memory plays a major role in modern digital systems due to its high-speed data access and efficient per formance. However, its reliable functioning is strongly influenced by device level parameters and supply voltage variations. In critical applications, even single fault occurrence may pose serious reliability issues, highlighting the need for ef ficient test methods. Extensive research has been carried out to investigate the static noise margin of SRAM cells. However, the influence of multiple defects has received relatively limited attention in existing literature. This study empha sizes the analysis of multiple defects because their occurrence becomes more fre quent in nano-meter technology regimes. Moreover, these defects can cause sig nificant fault behavior, potentially reducing the stability and reliability of SRAM cells. Multiple defects (Df3-Df3c) and (Df4-Df4c) are selected for analysis as they produce strong fault effects as they occur in the power supply and ground paths of the SRAM cell, which are critical for proper circuit operation. Any dis turbance along these conduction paths alters the effective operating voltage of the cross-coupled inverters and consequently affect the drive capability of the associated transistors. Moreover, the behavior of these defects is examined under various temperature conditions, supply voltages, and process corners in order to assess their overall effect on SRAM cell stability.
Authors - Sunil Jagannath Panchal, Gajanan Madhavrao Malwatkar Abstract - This research deals with the persistent challenges of document man agement in higher education institutions which focuses on the development of a digital support tool for Mariano Marcos State University (MMSU). Traditional paper-based systems and fragmented repositories often result in inefficiencies, duplication of work, and risks of data loss. The project adopted the Agile Devel opment methodology with emphasis on flexibility, collaboration, and iterative improvement. The d-T.R.A.I.L. system was built using JavaScript, PHP Laravel, HighCharts, and MySQL, integrating features such as tagging, repository man agement, granular access control, and collaborative modules like Teams. These functionalities were designed to streamline document organization, retrieval, and secure sharing across diverse academic and administrative units of the Univer sity. A User Acceptance Test (UAT) was conducted involving 70 participants from different MMSU offices that utilizes a Likert scale to measure satisfaction. Re sults yielded an overall mean score of 4.36 which was interpreted as Very Satis factory. High ratings were recorded for productivity, user-friendliness, and doc ument organization, while scalability received the lowest score which indicates an area for future enhancement of the system. The findings demonstrate that the system effectively improves workflow efficiency, accessibility, and accountabil ity, while aligning with national digital transformation policies.
Authors - Hileni Ihambo, Fungai Bhunu Shava, Gabriel Tuhafeni Nhinda Abstract - Fine-tuning large language models remains costly, and Parameter- Efficient Fine-Tuning (PEFT) techniques have emerged to make this process feasible on limited hardware. Among them, IA3 stands out for its extreme simplicity—it tunes only element-wise scaling vectors—but this design restricts the model to re-weighting features it already knows; it cannot form new ones. In this paper, we present SAMA (Spectral- Aware Minimal Adaptation), an extension of IA3 that introduces a single rank-1 update whose direction is derived from the pre-trained weights through Singular Value Decomposition. Each adapted layer gains only 4d extra parameters (3,072 for d=768), which is roughly one quarter of what LoRA requires at rank 8. We benchmark SAMA against five baselines—LoRA, DoRA, AdaLoRA, QLoRA, and IA3—across BERT, GPT-2, and Flan-T5 on twelve diverse NLP tasks under a low-resource constraint of 1,000 training samples per task. On the decoder-only GPT- 2, SAMA lowers perplexity by 26–34% relative to IA3 on both WikiText- 2 and Penn Treebank. On BERT’s RTE task, SAMA reaches 53.7% accuracy, surpassing IA3 (47.2%) and LoRA (52.6%) despite using 63% fewer trainable parameters than LoRA. We investigate this architecture dependence in detail and distil practical guidelines to help practitioners choose the right PEFT method for their setting.
Authors - M SANTHIYA, V KALAICHELVI Abstract - The wide use of machine learning in the field of medical imaging has caused concern with regard to patient information security, especially when mod els are being trained over multiple health care systems in a distributed manner. Centralized learning requires transferring raw patient data to a central server where there is an extreme risk of data breach and unauthorized access to patients' personal information. Violations of health care regulations (HIPAA and GDPR) can occur in a centralized system because of the transfer of patients' data. Feder ated Learning (FL) addresses these issues by allowing collaborative model de velopment on individual client devices. Therefore, the sensitive patient data will remain at its source institution. This paper provides a thorough comparative study of centralized learning and federated learning methods for detecting pneumonia utilizing chest X-rays from the publicly available Kaggle Chest X-Ray Pneumo nia dataset. Three architecture types (Support Vector Machine (SVM), Convolu tional Neural Network (CNN) and Long Short-Term Memory (LSTM)) were tested in both centralized and federated environments utilizing the FedAvg ag gregation method. Only the model weights were shared between the clients and the central server; therefore, patient data was maintained private through the en tire model training process. Experimental results demonstrated that federated learning produced superior performance than centralized learning for all three architectures (81.1%, 84.6%, and 82.7% for SVM, CNN and LSTM respec tively). The performance metrics for centralized learning were 76.6%, 76.3%, and 81.6%. This superior performance of FL is attributed to the inherent regular ization effect of local class-balancing within the federated clients that reduces the inherent class imbalance in the dataset. Overall, our research demonstrates that FL is not only a viable privacy-preserving solution to centralized training but offers improved generalization in the medical imaging domain with imbalanced classes and is a suitable solution for application in distributed health care envi ronments.
Authors - Vishruth B. Gowda, Sowmya T, Shreyas K, Megha J, Shreenidhi B S, Pranav Srinivas Abstract - Public administrations generate extensive administrative data through routine governance processes yet it is weakly based on verifiable evidence. This paper introduces a human-centric policy intelligence system based on execution-level administrative data for provision of accountable and evidence-based policy-making. The framework brings together governance-conscious data ingestion, cryptographic hash-based verification including permissioned blockchain systems to control the integrity of data, cross-domain data harmonisation to overcome administrative silos, and explainable machine learning models to create interpretable supporting insights. The framework is specifically meant as a human-in-the-loop system, maximizing policy foresight, administrative discretion, and accountability to the law. The validation with actual Mahatma Gandhi National Rural Employment Guarantee Act administrative data of the year 2022–2023 proves that the framework can be used to stress the implementation issues and regional inequalities without computerising policy-related decisions. The suggested solution is lightweight, scaled down to fit in the existing open-sector digital infrastructure.
Authors - Aprna Tripathi, Akhilesh Kumar Sharma, Avisikta Pal, Srikanth Prabhu, Ramakrishna Mundugar, Reet Ginotra Abstract - This paper presents a novel approach to identifying translation errors in Thai-English machine translation through the comparative analysis of multiple automatic evaluation metrics. Using a rank deviation methodology, we evaluate 350 Thai-English translations produced by seven contemporary systems provid ing online translations — including dedicated Machine Translation systems and large language models — across nine automatic evaluation metrics. By ranking translations within each metric and comparing individual metric rankings against the mean average rank, we identify translations that receive solitary punishment from a single metric, isolating these as candidates for manual error analysis. Our results demonstrate that individual metrics exhibit distinct sensitivity to specific error types, and that surface-level metrics retain diagnostic value along side advanced neural metrics. Neural metrics effectively identify meaning and adequacy errors, but surface-level metrics uniquely identify morphological vari ation, word order errors, preposition choice, and number formatting issues that neural metrics fail to penalize. The diversity of metric sensitivity is therefore an asset rather than an inconvenience, enabling a more complete characterization of translation error than any single metric can provide. This research supports the development of high-quality training data for MT fine-tuning by identifying the specific error types that individual metrics can and cannot detect and provides a repeatable diagnostic methodology applicable to other language pairs.
Authors - Bobby A. Eclarin, Mark Justine S. Cudapas Abstract - This research deals with the persistent challenges of document man agement in higher education institutions which focuses on the development of a digital support tool for Mariano Marcos State University (MMSU). Traditional paper-based systems and fragmented repositories often result in inefficiencies, duplication of work, and risks of data loss. The project adopted the Agile Devel opment methodology with emphasis on flexibility, collaboration, and iterative improvement. The d-T.R.A.I.L. system was built using JavaScript, PHP Laravel, HighCharts, and MySQL, integrating features such as tagging, repository man agement, granular access control, and collaborative modules like Teams. These functionalities were designed to streamline document organization, retrieval, and secure sharing across diverse academic and administrative units of the Univer sity. A User Acceptance Test (UAT) was conducted involving 70 participants from different MMSU offices that utilizes a Likert scale to measure satisfaction. Re sults yielded an overall mean score of 4.36 which was interpreted as Very Satis factory. High ratings were recorded for productivity, user-friendliness, and doc ument organization, while scalability received the lowest score which indicates an area for future enhancement of the system. The findings demonstrate that the system effectively improves workflow efficiency, accessibility, and accountabil ity, while aligning with national digital transformation policies.
Authors - Gauthaman S P, Paneer Thanu Swaroop C, Bagavathi Sivakumar P, Anantha Narayanan V Abstract - Psoriasis is a long-term inflammatory skin disease commonly identi fied by red plaques, scaling, and abnormal thickening of the epidermis. Reliable evaluation of disease severity is important for determining appropriate treatment options and for tracking patient response to therapy. In clinical practice, severity is often assessed using the Psoriasis Area and Severity Index (PASI). Although widely adopted, this method largely depends on visual examination and clinician judgment, which may lead to inconsistencies and observer-dependent variations. Recent developments in artificial intelligence and non-invasive dermatological imaging technologies provide opportunities for more objective and automated assessment of skin disorders. In this study, a novel framework is proposed for psoriasis severity evaluation that integrates skin biomechanical characteristics with deep hybrid learning mod els. Biomechanical attributes of the skin, including elasticity, stiffness, and vis coelastic behavior, are obtained through non-invasive measurement techniques and combined with visual information derived from dermatological images. The proposed system employs a hybrid deep learning architecture that incorporates convolutional neural networks (CNN) for image feature extraction along with machine learning classifiers for severity prediction. By jointly analyzing biome chanical and visual features, the framework aims to enhance the precision, con sistency, and reproducibility of psoriasis severity assessment. Experimental anal ysis indicates that the inclusion of biomechanical biomarkers alongside deep learning significantly improves prediction performance when compared with tra ditional image-based models. The proposed approach can support dermatologists in clinical decision-making and may also facilitate applications in tele-dermatol ogy and personalized disease monitoring.
Authors - Vijayanirmala Baddala, Jolakula Asoka Smitha, Bichagal Shadaksharappa Abstract - Accurate State-of-Charge (SoC) estimation is critical for ensuring the reliability, safety, and operational efficiency of lithium-ion batteries in electric vehicles and energy storage systems. While data-driven models offer high precision, centralized approaches are increasingly limited by data privacy concerns, high communi- cation overhead, and poor scalability. This paper addresses these challenges by proposing a comprehensive deep learning and federated learning (FL) frame- work for decentralized SoC prediction using the OSF battery dataset. We use four LSTM architectures: Stacked LSTM, Bidirectional LSTM, Attention-based LSTM, and Stateful LSTM, which are integrated into a federated model to sys- tematically evaluate their performance. These include FedAvg, FedProx, and adaptive methods such as FedAdam and FedYogi. To our knowledge, this is the first study to evaluate these architectures in the context of a federated battery management system (BMS). Results show that The comparative analysis inves- tigates the interplay between model complexity and federated optimization, with a specific focus on predictive accuracy, convergence behavior, and robustness to non-IID data distributions stemming from heterogeneous battery capacities and usage patterns. By benchmarking these combinations, this research identifies optimal strategies for implementing privacy-preserving, communication-efficient, and scalable Battery Management Systems (BMS) at the edge.
Authors - Abir Paul, Priti Giri, Rajdeep Ghatak, Soumitra Sasmal, Mauparna Nandan, Partho Mallick Abstract - Accurate forecasting of drug demand is one of the challenging areas in the healthcare service to reduce waste as well as shortages. Some recent studies focused only on predicting drug use demand for regions and hospitals, missing an overall way to combine these forecasts. In this study, a multilevel machine learning framework is presented that merges regional tender demand predictions with monthly and seasonal order forecasting in hospitals and pharmacies. With historical drug usage, the system captures time-based changes, seasonal demands, and also location specific behaviors . Models for regional tenders predict yearly procurement, but models at hospitals and pharmacies try to tell the need of each month, allowing better resource distribution.The rigorous experimental process showed better estimates and forecasting with less error than just making a single-level prediction. This framework helps to make better purchasing decisions and ensures a stable drug supply across healthcare systems. Health departments, hospital chains, and pharmacy groups can benefit from using a model .
Authors - Gagandeep Malhotra, Dharm Singh Jat Abstract - Modern Electronic Warfare (EW) environments are very dynamic, crowded, and hostile, which makes static or centralised spectrum-allocation strategies useless. To tackle these issues, this paper introduces a completely adaptable Federated Deep Q-Network (A-FDQN) framework for each node, which is built onto a high-fidelity ns-3.40 EW simulation environment. In this simulation each tactical radio has been configured to work as an independent federated client which trains a local DQN within itself based on metrics obtained from SINR, congestion, delay, jitter, and interference caused by jamming. A federated server then periodically collects client models using Federated Averaging (FedAvg) or Median method. This lets global learning happen without needing centralised state visibility or constant connectivity, which is very important for networks on contested battlefields. Our framework is different from earlier RL and FL studies because it combines per-node reinforcement learning, periodic federated aggregation, partial client participation, tuneable synchronisation frequency, and realistic ns-3 modelling of mobility, sweep jamming, bursty traffic, congestion hotspots, and Wi-Fi PHY/MAC interactions. Our A-FDQN system dynamically changes the channel assignments at each node, which greatly improves the packet delivery ratio, minimum SINR, fairness, and delay when faced with challenging EW scenarios. This first of its kind end-to-end FRL architecture offers a decentralised, resilient, and operationally relevant way to manage the spectrum for next-generation military communications.
Authors - Harsh Vardhan, Harsh Vikramaditya, Doyelshree Bhui, Shilpi Basak, Soumitra Sasmal, Subhajit Bhowmick, Ishan Ghosh Abstract - Security audits present a unique and ever evolving challenge due to the dynamic nature of cyberthreats and complex regulations. Traditional compliance audits remain largely manual and labor inten sive, resulting in vast inconsistencies. This paper introduces a solution to make compliance audits easier and faster by proposing a framework that leverages the use of Natural Language Processing and Large Lan guage Models to map organizational policies to frameworks and allows for real-time data from security controls to be validated against these complex security frameworks. Through a hybrid multi-model architec ture, the solutions in this paper aim to enhance the accuracy and trans parency of compliance evaluations coupled with evidence-backed insights. The results demonstrate the potential of integrating intelligent auditing systems to deliver compliance assessments that are consistent, accurate, and rapid; streamlining governance and improving cyber security posture management.
Authors - Anjali Yawatkar, Hemlata Gaikwad Abstract - Contemporary customer support systems require processing a massive number of user queries with low latency and high semantic relevance. Rule-based systems fail to capture context, while fully LLM-based systems are computation ally expensive and suffer from high latency. This paper introduces an adaptive AI-assisted customer support automation system using an optimized Retrieval Augmented Generation (RAG) model. The proposed system combines Azure OpenAI embeddings, FAISS-based vector search, selective Cross-Encoder re ranking, and a Learning-to-Rank (LambdaMART) model for adaptive score fu sion. Unlike vanilla RAG models, the proposed system adaptively re-ranks only the top-k retrieved candidates, trading off ranking precision and latency. Experi ments were carried out on a 1,30,000-sample e-commerce customer support da taset with query-response pairs annotated with intent labels. Compared to rule based retrieval, embedding+FAISS, and vanilla RAG models, the proposed hy brid system showed improved top-1 retrieval precision with a concurrent reduc tion in end-to-end latency from 0.414s to 0.365s (≈11.8% relative improvement). The LambdaMART model adaptively learned weights from FAISS and Cross Encoder scores, improving ranking robustness and eliminating misranked top re sponses. The system was implemented on Azure Machine Learning with a cloud scale pipeline and interactive Streamlit web interface, showcasing the cost-effec tive inference capabilities of the proposed system via selective re-ranking.
Authors - Jaykumar Gandharva, Hardika Menghani, Tilak Brahmbhatt, Nischay Agrawal Abstract - Modern Electronic Warfare (EW) environments are very dynamic, crowded, and hostile, which makes static or centralised spectrum-allocation strategies useless. To tackle these issues, this paper introduces a completely adaptable Federated Deep Q-Network (A-FDQN) framework for each node, which is built onto a high-fidelity ns-3.40 EW simulation environment. In this simulation each tactical radio has been configured to work as an independent federated client which trains a local DQN within itself based on metrics obtained from SINR, congestion, delay, jitter, and interference caused by jamming. A federated server then periodically collects client models using Federated Averaging (FedAvg) or Median method. This lets global learning happen without needing centralised state visibility or constant connectivity, which is very important for networks on contested battlefields. Our framework is different from earlier RL and FL studies because it combines per-node reinforcement learning, periodic federated aggregation, partial client participation, tuneable synchronisation frequency, and realistic ns-3 modelling of mobility, sweep jamming, bursty traffic, congestion hotspots, and Wi-Fi PHY/MAC interactions. Our A-FDQN system dynamically changes the channel assignments at each node, which greatly improves the packet delivery ratio, minimum SINR, fairness, and delay when faced with challenging EW scenarios. This first of its kind end-to-end FRL architecture offers a decentralised, resilient, and operationally relevant way to manage the spectrum for next-generation military communications.
Authors - Sachin Kumar Abstract - E-commerce search engines rely on Query Expansion (QE) to bridge the semantic gap between user queries and product catalogs, but expansion can induce query drift, where retrieved results diverge from the user’s original intent. Evaluating QE on novel or out-of-distribution queries is fundamentally intractable under the standard Cranfield paradigm, which requires pre-compiled relevance judgments. This paper introduces the Generalized Authority-Hub Score (GAHS), an unsupervised evaluation metric that repurposes the product catalog’s relational structure— modeled as a product graph—as a dynamic proxy for retrieval quality. Drawing on the HITS algorithm, GAHS quantifies the topical coherence of a retrieved product set without requiring explicit relevance judgments. Using the Amazon ESCI dataset, we validate GAHS against MAP and nDCG@10 on a held-out seen query set, demonstrating strong rank-order agreement (Kendall’s τ = 1.0 with MAP, τ = 0.67 with nDCG@10). We further demonstrate its discriminative power on a disjoint unseen query set, and discuss an observed performance reversal between the two query sets and its implications for QE evaluation methodology.
Authors - K Devi Priya, P Saranya Durga, Y Sony, D Varun Sai Abstract - This paper presents a comprehensive implementation and evaluation of a secure electronic voting system built on the Ethereum blockchain platform. Proposing on Ethereum smart contracts, Proof of Stake consensus, and modern Web3 technologies and implemented the project. The implementation deals with key e-voting issues like voter authentication, ballot privacy, vote immutability and transparent auditability.We examine security threats, offer Layer2 scaling design, introduce concepts of zero-knowledge proofs in order to achieve higher privacy levels, and measure the economic benefit of deployment on different scales. In our results, we have shown that Ethereum has a significant basis to support decentralized voting systems, but scalability and cost reduction remain an important challenge to large-scale elections. The paper ends with a set of practical recommendations on the deployment of production and the main directions on the further research in the field of blockchain-based democratic systems.
Authors - Manav Thakar, Nischay Agrawal, Jaykumar Gandharva, Manish Singh Abstract - Predicting and understanding the inhibitory activity associated with Breast Cancer resistance protein can assist in the drug discovery process by anticipating the potential drug resistance and drug-drug interactions. Prediction of BCRP inhibitors using machine learning can accelerate the identification of BCRP inhibitors by analyzing large datasets, finding patterns in molecular structures, and predicting interactions that would be time-consuming and expensive through traditional methods like high-throughput screening or trial-and-error experimentation. In the literature, machine learning has been employed to develop techniques for predicting BCRP inhibition. However, these methods often exhibit low prediction accuracy, highlighting the need for improved prediction techniques with enhanced accuracy. In this research, BCRP inhibition prediction has been carried out using features spaces fusion to enhance the features information with richer representation of data incorporating complementary aspects of molecule to get the increased accuracy for discovery of inhibitors for drugs of breast cancer. The experimental results show that the proposed technique has increased accuracy and precision for the discovery of BCRP inhibitors. The accuracy of the proposed technique is 97% which is higher than the techniques developed in literature. The study demonstrates that enhancing the features information by combining various compound properties creates a more richer and comprehensive feature space. This enhanced feature representation can significantly help in identifying BCRP inhibitors specifically and contribute to advancements in drug discovery overall.
Authors - Shaik Sohail Ahammed, B. Rohan Teja, R. Naga Sumithra, D. Manasa, T. N. V. D. Sai Krishna Abstract - Privilege Escalation is a major issue for securing Linux sys tems. When a user gains unauthorized root access he has the ability to access all system resources and manipulate them at will. In the past, Linux has used Static Access Control Policies and User Space Monitoring Tools to secure system access. However, these methods provide little in sight into how the kernel is modifying users credentials when permissions are changed. In this paper we propose a Kernel-Level solution to detect and prevent unauthorized privilege escalations. This detection/ preven tion occurs in real time via a Credential Transition Monitoring Mecha nism within the kernel layer, which prevents the elevation of privileges by illegal means. To create the functionality necessary for the above, a Linux Kernel Module (LKM) was created which utilizes kprobes to in tercept calls to the commit creds() function, which is used to update a processes credentials in the kernel. To evaluate if the privilege escalation being requested is legitimate or malicious, the LKM contains a Policy Based Evaluation Mechanism which evaluates each request to modify a process’s credentials. We tested our proposed solution using a con trolled test environment composed of a Virtual Machine (VM) running the Ubuntu Operating System. We ran two types of tests, first were Le gitimate Administrative Operations utilizing the ”sudo” utility, second were Simulated Privilege Escalation Attacks based upon SetUID Vul nerabilities. Our results show that the system effectively detected and blocked malicious privilege escalations, while providing minimal over head to normal system operation.
Authors - Menna Elgabry, Ali Hamdi Abstract - Mortality prediction for intensive care unit (ICU) patients with alcohol-related disorders remains insufficiently explored despite the distinct clinical characteristics and elevated risk profile of this population. Unlike general ICU cohorts, these patients often present with impaired physiological function, frequent complications, and poorer overall outcomes. However, few research works have taken this patient group into account for mortality prediction. This study addresses the gap by developing mortality prediction models specifically for ICU patients with alcohol-related disorders using multimodal electronic health record data. To capture the complex clinical status of patients, we integrate six major data modalities in the first 24 hours after admission, including demographics, diagnoses, medications, procedures, laboratory results/vital signs, and patient outputs. A refined preprocessing pipeline was used to harmonize and process heterogeneous input data. In addition, severe class imbalance is another challenging issue in resolving this mortality predict task. Therefore, our work examines systematically several rebalancing strategies: no resampling, oversampling, undersampling, and SMOTENC. Evaluated on both MIMIC-III and MIMIC-IV databases, our proposed rebalanced multimodal data approach is effective for tackling the task. Indeed, the experimental results show that CatBoost with random undersampling provides the most consistent and balanced effectiveness. Furthermore, multimodal analysis demonstrates that combining diagnoses, laboratory results/vital signs, and medications substantially improves prediction, while integrating all modalities achieves the best overall performance.
