Surendra Solanki

An accurate crop yield prediction is necessary for augmenting farming practices and guaranteeing the security of food. Several approaches have been developed to predict the yield of crops. The conventional approaches were incapable of discovering non-linear relationship amid the data, and they exhibit several drawbacks, such as computational overhead, false errors, overfitting issues, limited resources, poor generalizability and computational complexity, which lead to inaccurate predictions. To overcome these limitations, the research proposed a Multi-Attention and Distilled Task Learning enabled Weighted Decision tree based Bidirectional Long Short-Term Memory (MA-DTL-WDBiLSTM) model to predict the crop yield accurately. The model’s performance in forecasting the crop yield is enhanced due to the incorporation of multi attention mechanism and Distilled Task Learning, which distilled the gained knowledge from the teacher model to the student model and split the tasks into subtasks, respectively. Additionally, the integration of the weighted decision tree in the Bi-LSTM model offers an improved prediction accuracy by handling the intricate data and concentrating on the relevant features. The model is evaluated using the Crop yield prediction dataset and the Agricultural Crop Yield in Indian States Dataset, which enhances the interpretability and better generalizability. Specifically, the proposed MA-DTL-WDBiLSTM model achieved a minimum correlation of 0.884, MAE of 1.759, MSE of 6.536, and RMSE of 2.556 using the Agricultural Crop Yield in Indian States Dataset, showcasing high performance in crop yield prediction and outperforming other existing models.

Advanced malware investigations frequently extend over prolonged periods, during which operating-system updates, software patches, logging-framework modifications, benign software evolution, and adversarial adaptations continuously alter the statistical characteristics of digital forensic artifacts. These temporal changes introduce evidence drift that affects registry traces, process lineage structures, executable metadata, memory layouts, network flows, and system-call distributions, thereby reducing the reliability and long-term validity of forensic evidence. Although existing concept-drift approaches improve predictive performance under evolving data distributions, they rarely assess whether forensic artifacts remain evidentially valid, reproducible, and operationally reliable throughout extended investigations. To address this limitation, this paper proposes a five-stage Evidence Drift Analytics (EDA) framework for long-term forensic-validity optimization in evolving malware environments. The framework integrates the Temporal Forensic Evidence Divergence Field (TFEDF) for multi-resolution drift estimation, the Causal Artifact Lineage Graph Transformer (CALGT) for SCM-guided estimation of attribution confidence under explicit intervention assumptions, the Forensic Integrity Preservation Index using Bayesian Evidence Survival Modeling (FIPI-BESM) for evidentiary survivability forecasting, the Drift-Aware Counterfactual Re-Anchoring Engine (DACRE) for learning drift-invariant representations while preserving malicious behavioural semantics, and the Forensic Validity Optimization Network (FVON) for reinforcement-learning–based forensic decision optimization. The proposed framework was evaluated using five publicly available cybersecurity datasets, including CICIDS2017, EMBER, LANL authentication logs, Android API streams, and IoT telemetry datasets. Experimental results demonstrate that the proposed approach achieves 94.8% drift sensitivity with a 4.2% false-positive rate, attains 91.7% accuracy for the operational drift-source attribution task under the stated SCM intervention assumptions, and maintains a survival stability index of 0.92 during long-term evidentiary forecasting. Furthermore, the proposed drift-aware representation learning reduces embedding variance to 0.062 while preserving 96.1% reconstruction fidelity across temporal windows. The integrated framework also achieves a cumulative forensic validity score of 0.95 and a reproducibility index of 0.94, indicating improved cross-temporal consistency, evidentiary robustness, and operational reliability under evolving cybersecurity environments. Overall, the proposed Evidence Drift Analytics framework provides a unified evidence-centric methodology for modelling temporal evidence drift, preserving forensic validity, and supporting reliable long-term malware investigations through the integration of temporal drift analysis, SCM-guided causal reasoning, Bayesian evidence survival modelling, drift-invariant representation learning, and reinforcement-learning–based forensic optimization. These performance results were obtained under the offline experimental settings evaluated in this study using temporally segmented benchmark datasets and may vary under real-time, high-throughput operational environments. The CALGT outputs represent estimated attribution confidence obtained under assumptions of temporal ordering, reliable metadata observability, and correctly identified operating-system update pathways; they should not be interpreted as definitive causal discovery or experimental proof of causality. .

Sixth generation (6G) networks require intelligent and privacy-preserving solutions to handle traffic complexity, resource allocation, and security challenges. While Large Language Models (LLMs) offer advanced reasoning and Federated Artificial Intelligence (FAI) enables distributed learning without directly sharing raw data, their combined application in large-scale network science remains limited. Key issues include non-IID data, decision latency, and limited robustness of centralized models. To address these challenges, this work introduces FALCON-6G (Federated AI and LLM-Oriented Cognitive Optimization in 6G Networks), a novel framework for distributed anomaly detection, traffic prediction, and adaptive resource management. Experiments conducted under federated learning settings with 50 participating clients and 200 global communication rounds on two Open RAN and four intrusion detection datasets show a 22.7% average relative improvement in throughput prediction accuracy and a 17.6% reduction in decision latency compared with FedAvg, along with up to a 23.9% relative improvement in intrusion detection F1-score over FedAvg and a 34.0% reduction in communication overhead. These results demonstrate the potential of FALCON-6G for privacy-aware 6G network optimization. The current evaluation is primarily based on benchmark datasets, and validation in live large-scale 6G/O-RAN environments remains an important direction for future work.

The increasing sophistication and polymorphic nature of malware necessitate intelligent, adaptive forensic frameworks that go beyond static rule-based detection. In particular, the traditional techniques for evidence collection and analysis would not catch the evolving behaviors of advanced malware, lack context-aware feature weighting, and fail to maintain accuracy under concept drifts or temporal drifts. Such limitations thus also impede timely threats attribution and malware family discovery processes. To bridge these gaps, we propose an iterative framework for evidence collection and malware activity analysis by utilizing machine learning and explainable AI paradigms. The framework integrates five newly proposed modules: (1) DynaTrace-XAI employs SHAP-based attribution for real-time reweighting of forensic features, dynamically enhancing the significance of impactful indicators in the process. (2) TempNet V introduces temporal ensemble validation at daily, weekly, and monthly levels, deploying heterogeneous models (e.g. LSTM, RF) to counter data drifts dependent on time. (3) GRAFIN-A reconstructs execution traces into behavior graphs through Graph Attention Networks, which are then subjected to unsupervised clustering with DBSCAN and anomaly detection through graph similarity scoring methods. (4) RiskActive-L blends anomaly-derived risk scores with model uncertainty to prioritize more unlabeled high-risk samples within an active learning loop, strongly improving data labeling efficiency. (5) DriftMon-F implements a forensic-context-aware drift detection mechanism via KL divergence and meta-log correlation to trigger timely retraining in the event of contextual shifts. In concert, these modules constitute a closed-loop system whereby evidence assignment, model validation, clustering, sample selection, and drift handling processes become continuously refined. Empirical evaluations demonstrate the empirical achievements: 15% increase in F1-score, 23% gain in variant detection, reduction of 45% in labeling effort, and 70% reduction in drift recovery lag, leading to considerable advances in adaptive malware forensics.

Cyber-forensic investigations increasingly involve heterogeneous evidence, including phishing audio, malicious text and scripts, executable binaries, memory artifacts, and network traffic. Existing approaches generally analyze these sources independently or perform sample-level fusion without explicitly addressing temporal asynchrony, higher-order forensic relationships, missing attack stages, counterfactual threat evolution, and attribution uncertainty. This paper presents a Multimodal Generative Cyber Forensics Network comprising CTEEN for temporal-semantic alignment, NSFRH for higher-order forensic reasoning, GASRE for evidence-constrained attack-stage reconstruction, CMES for counterfactual malware-evolution analysis, and CTAECO for confidence-aware candidate-category attribution. The evaluation used VirusShare, CIC-MalMem-2022, the Android Malware Corpus, the Malware Traffic Repository, and an Integrated Controlled Multimodal Benchmark constructed from source pools containing approximately 52,000 malware records, 1.8 million network-flow records, 24,000 textual or script artifacts, and 12,500 phishing or social-engineering audio recordings. Comparisons were conducted against Cyber Code Intelligence, AI-Based Malware Analysis, Multimodal Pre-trained Fusion, and LLM-MalDetect using identical eligible evidence groups, preprocessing outputs, partitions, and metric definitions. Across 25 matched partition–seed observations, the proposed framework achieved 99.26 ± 0.18% classification accuracy (95% CI: 99.19–99.33%), 99.20 ± 0.21% F1-score, and 99.66 ± 0.12% AUC. It further obtained 98.46% mean controlled alignment accuracy, 98.68% evidence-constrained reconstruction accuracy, 99.04% benchmark-defined candidate-category attribution accuracy, and 99.0% overall forensic-intelligence reliability. Accuracy decreased to 95.18% under temporally disjoint testing and 92.34% under family-disjoint testing. Leakage audits, negative controls, matched statistical tests, and alternative-design ablations further supported the findings. The complete framework contained 96.3 million trainable parameters and required 12.8 GB of peak memory, with a mean inference latency of 73.2 ms per complete multimodal episode (13.7 episodes/s) under the reported computational environment. These requirements indicate that deployment on resource-constrained platforms would require model compression or inference optimization. Reviewer 3, Comment 3) The results demonstrate effective evidence-constrained multimodal reasoning under the controlled benchmark but do not establish natural incident-level correspondence, definitive real-world attribution, or universal forensic performance.

Metaheuristic optimization algorithms inspired by animal behaviors have predominantly focused on wild and predatory species, while the structured and cooperative behaviors of domesticated animals remain largely unexplored. Motivated by this gap, this paper proposes a unified family of four domesticated animal–inspired metaheuristic algorithms: Canine–Human Trust Optimization (CHTO), Feline Curiosity Optimization (FCO), Equine–Human Harmony Optimization (EHHO), and Lagomorph Bond Optimization (LBO), where lagomorph refers to rabbit-inspired bonding behavior. Each algorithm models companionship-oriented traits such as trust, curiosity, harmony, and social bonding to achieve an effective balance between exploration and exploitation. The proposed algorithms are evaluated on standard unimodal and multimodal benchmark functions and dynamic optimization problems using the Generalized Moving Peaks Benchmark (GMPB). Experimental results show that EHHO consistently outperforms classical and recent metaheuristics, including Particle Swarm Optimization (PSO), Harris Hawks Optimization (HHO), Slime Mould Algorithm (SMA), and Aquila Optimizer (AO). On highly multimodal benchmark functions such as Rastrigin (f₃), Griewank (f₄), and Levy (f₇), EHHO achieves up to 35% improvement in mean fitness values compared to Harris Hawks Optimization (HHO), which is identified as the strongest competing baseline algorithm. In dynamic optimization scenarios using the GMPB, EHHO reduces the average offline error by approximately 28% compared to Harris Hawks Optimization (HHO), while achieving a success rate of 92% across changing environments. Statistical validation using Friedman and Wilcoxon signed-rank tests confirms the robustness and statistical significance of the proposed algorithms. The findings establish domesticated animal companionship behaviors as a competitive and effective paradigm for solving both static and dynamic optimization problems.

Automatic image captioning aims to acquire the learning of the correlation between the visual and the linguistic cognitive domain through the generation of descriptive captions with respect to input images. But it is affected by the semantic distance between image features and text, and problems in modeling fine-grained object interactions and contexts. We present a new encoder-decoder architecture, that utilizes optimal vision transformer + attention to the transformer by using cosine distance in order to get more context by providing better mapping. It relies on Inception-V3 to acquire fine-grained visual features and skip-gram features of the textual semantics. Another algorithm named improved walrus optimization dynamically updates the key hyperparameters to enhance convergence and performance. Experiments on the MS-COCO and Flickr8k datasets show that it is better than prior methods with BLEU-4 scores of 0.8864 and 0.8920, and higher performance gains in ROUGE, METEOR and CIDEr scores. In conclusion, the method outputs with better contextual details and flow of natural language, semantic fidelity. It is optimal in practice due to its cosine attention and adaptive tuning.

Alcoholism, habitual and excessive intake of alcoholic beverages, presents a significant disorder that challenges contemporary society; however, alcoholism detection lacks universally acknowledged examinations or protocols. Traditional subjective methods are time-consuming and prone to error. Electroencephalography (EEG) detects alcoholism by analyzing the electrical activity of the brain. An effective EEG channel selection using metaheuristic algorithms (MHAs)-based features are introduced. The EEG signal is decomposed into subbands using a novel optimal wavelet filter bank (OWFB). Each subband is represented using four features: mean, Higuchi’s fractal dimension, log entropy, and Rényi’s entropy. The optimal subband features are investigated using combinations of four MHAs and are six classification models. A publicly available 64-channel EEG dataset of alcoholic and non-alcoholic signals is employed. Among the evaluated combinations, Sparrow search algorithm combined with k-nearest neighbor model (KNN) classifier achieved the highest accuracy of 95.90% and F1-score of 96.80%, closely comparable to using all EEG channels (96.30% accuracy and 96.83% F1-score). Overall, KNN classifier consistently outperformed others, indicating that optimal channel selection can effectively reduce channel redundancy while maintaining high accuracy in alcoholism detection. The optimal channel selection enhances the performance of the ML models and reduces the computational time compared to existing alcoholism detection methods. A Python implementation is available at https://github.com/pramodkachare/EEG_Optimal_Wavelet_MHA .

The growing adoption of electric vehicles (EVs) demands reliable and economically viable charging infrastructures that minimize investment risk while ensuring grid stability. Conventional planning approaches often fall short by either overestimating installation needs or underutilizing existing network capacity. This study presents an Entangled Adaptive Hybrid Quantum-Inspired NSGA-II (EAH-QNSGA-II) designed to address these challenges through a multi-objective optimization framework. The method incorporates quantum-inspired representation, entanglement-driven crossover, adaptive quantum rotation, and a localized search mechanism, enabling a balanced exploration–exploitation process and maintaining high-quality Pareto solutions. The proposed framework is evaluated on four benchmark datasets: Palo Alto EV charging records, Boulder public charging usage, the Multi-Faceted EV Charging Transactions dataset, and the IEEE 33-bus distribution system for grid integration. The proposed method aims at minimizing installation cost while simultaneously maximizing service coverage and improving grid load stability in the presence of renewable energy variability. Experimental findings reveal that EAH-QNSGA-II delivers significant gains compared with state-of-the-art approaches. Against classical NSGA-II, the method achieves 34.1% lower installation cost, 30.2% better coverage, and 41.7% stronger grid load balancing. When compared with quantum-inspired PSO, Jaya, EAQGA, and AHQSOA, the proposed method enhances hypervolume by 18–24% and spread by 15–19%, highlighting superior convergence and diversity. These results confirm EAH-QNSGA-II as an efficient and scalable solution framework for future EV charging infrastructure deployment.