Category Archives: Uncategorized

Authors: Sagar Gupta

Abstract: Crowdsourcing has emerged as a powerful mechanism for harnessing distributed human intelligence at scale, enabling diverse applications such as data annotation, collective problem solving, and decision-making across domains. With the advent of neural networks, crowdsourced activity has been both a source of critical training data and an arena for deploying advanced artificial intelligence systems to optimize participation, reliability, and outcome quality. This paper explores the intersection between crowdsourced activity and neural networks, emphasizing how neural architectures are applied to classify, validate, and enhance crowd contributions. The discussion spans natural language processing, computer vision, recommendation systems, quality assurance, and hybrid human–AI collaboration frameworks. The review concludes with challenges in scalability, bias mitigation, and ethical considerations, highlighting emerging opportunities for integrating neural networks to reshape crowdsourced ecosystems.

DOI: http://doi.org/10.5281/zenodo.17075867

Published by: vikaspatanker

Authors: Sagar Gupta

Abstract: Enterprise Resource Planning (ERP) systems have long served as the backbone of organizational information systems, integrating finance, operations, human resources, supply chains, and customer-facing processes into unified platforms. Traditionally, ERP implementations relied on rule-based configurations and deterministic workflows. However, the evolution of neural networks has introduced adaptive, data-driven intelligence into ERP ecosystems. Neural architectures are increasingly being deployed to enhance demand forecasting, anomaly detection, process optimization, and user personalization within ERP systems. This paper traces the evolution of neural networks in ERP implementations, from early adoption in predictive analytics to contemporary applications in autonomous process automation and decision intelligence. It also explores case studies, challenges, and future research directions, highlighting the transformative potential of neural networks in reshaping the ERP landscape

DOI: http://doi.org/10.5281/zenodo.17075766

Published by: vikaspatanker

Authors: Sagar Gupta

Abstract: The financial domain is inherently dynamic, stochastic, and complex, making it one of the most fertile grounds for the application of advanced machine learning techniques. Among these, Recurrent Neural Networks (RNNs) have emerged as particularly well-suited for modeling sequential and temporal dependencies in financial data. This paper explores the role of RNNs in complex finance applications, tracing their evolution from basic time-series forecasting to modern variants such as Long Short-Term Memory (LSTM) networks and Gated Recurrent Units (GRUs). The discussion highlights applications in algorithmic trading, credit risk assessment, fraud detection, portfolio optimization, and regulatory compliance. Case studies are presented to illustrate both the potential and the limitations of RNNs in finance. The paper concludes with a critical discussion of challenges such as interpretability, overfitting, adversarial risks, and future research directions, including hybrid neuro-symbolic architectures and transformer-RNN hybrids for financial intelligence

DOI: http://doi.org/10.5281/zenodo.17075673

Published by: vikaspatanker

Authors: Sagar Gupta

Abstract: Recent advances in Natural Language Processing (NLP) have been driven by the widespread adoption of large-scale pretrained language models (LMs). While generic NLP models such as GPT, BERT, and T5 exhibit strong zero-shot and few-shot performance across diverse tasks, specialized NLP models (e.g., BioBERT, FinBERT, SciBERT) are fine-tuned on domain-specific corpora to achieve superior performance in targeted applications. With the emergence of prompt engineering as a method to guide large language models (LLMs), a new research challenge arises: can prompt engineering narrow the performance gap between generic and specialized models, or does domain-specific pretraining remain necessary? This paper provides a comparative analysis of generic and specialized NLP models under different prompt-engineering strategies, focusing on domains such as finance, healthcare, and legal text processing. Experimental findings indicate that while prompt engineering enhances the adaptability of generic LMs, specialized models continue to outperform in precision-critical tasks. The study underscores the complementary role of prompt design and domain-specific adaptation in the next generation of NLP systems

DOI: http://doi.org/10.5281/zenodo.17075611

Published by: vikaspatanker

Authors: Noor jahan Fatima, Dr sarabjit kaur

Abstract: Graceful labelling is a fundamental problem in graph theory with significant applications in communication networks, coding theory, VLSI circuit design, and combinatorial optimization. The graceful tree conjecture, proposed by Rosa in 1967, asserts that every tree can be assigned a graceful labelling, yet a general proof or counterexample remains elusive. Traditional constructive techniques and exhaustive searches have established results for specific graph families, but scalability challenges persist when dealing with larger instances. This paper investigates scalable and efficient approaches to graceful labelling by integrating constructive methods with heuristic and optimization-based strategies. We explore hybrid approaches that combine deterministic recursive labelling with stochastic metaheuristics such as genetic algorithms, simulated annealing, and tabu search. Additionally, we examine the role of integer linear programming (ILP), constraint satisfaction formulations, and parallel algorithms leveraging GPU acceleration and distributed computing frameworks. Experimental evaluations demonstrate that hybrid and parallel approaches outperform traditional heuristics in terms of scalability and efficiency, particularly for large trees and special graph families. Approximation-based relaxations are also shown to provide near-graceful solutions that guide heuristic refinement. Beyond computational advancements, this work highlights theoretical implications, including potential structural insights into the graceful tree conjecture and extensions to cyclic graphs. The proposed scalable frameworks not only advance computational verification but also contribute to bridging the gap between practical applications and theoretical challenges in graph labelling.

DOI:

 

 

Published by: vikaspatanker

Authors: Chilufya Sydney

Abstract: This paper introduces an open-source, simplified facial recognition system intended to prevent impersonation during Zambia’s GCE exams. The system uses Python, OpenCV, and SQLite to perform two basic functions: enrolling student facial data at pre-exam registration and verifying identities in real-time at exam entrance. With a resource-poor environment perspective, the solution is affordable (<$100 per exam site equivalent to about K3,000 in Zambian currency), simple to deploy, and ethically deployed. The initiative fills significant literatures gaps concerning the flexibility of facial recognition systems in sub-Saharan African testing regimes and ethical adoption of biometric technology in schooling. Analysis of technical operational performance happens under normal lighting and equipment use through dataset-based controlled benchmarking and usability estimates rely on stakeholder evaluations. Under perfect laboratory settings, the system achieves 96.4% accuracy while delivering 87.8% accuracy at low luminance levels using 1.45 seconds for each verification process. The thesis describes the entire system development lifecycle from its inception until assessment while presenting actual field results. The system presents a practical application to enhance test security within educational institutions of developing regions. The solution delivers operational tractability alongside technical sophistication by giving an applicable solution to combat examination impersonation without exceeding existing resource capabilities.

DOI: http://doi.org/

Published by: vikaspatanker

Authors: Praveen Kumar Yadav

Abstract: Monitoring and maintaining water quality in urban rivers is crucial for ensuring both environmental sustainability and public health. The Gomti River, which flows through the densely populated city of Lucknow, faces severe stress due to rapid urbanization, untreated wastewater discharge, and growing anthropogenic pressures. This study focuses on predicting three key water quality parameters—pH, nitrate (NO₃), and biochemical oxygen demand (BOD)—which are widely recognized as critical indicators of river water health and are frequently used in Water Quality Index (WQI) assessment. To achieve this, the Extra Trees Regressor (ETR) model was applied to water quality datasets collected from the Central Water Commission monitoring station between 2016 and 2023. The dataset was pre-processed, normalized, and divided into training and testing subsets. Model performance was evaluated using statistical metrics such as R², MAE, MSE, and RMSE. The results demonstrated that ETR provided highly accurate predictions, achieving R² values above 0.95 for all three parameters while minimizing error values (MAE and RMSE). The predicted WQI values showed close alignment with actual observations, confirming the robustness and reliability of the model. These findings highlight the potential of machine learning-based approaches in forecasting river water quality and supporting timely, data-driven decision-making for pollution control and river management. This research contributes to environmental monitoring and geoscientific applications by demonstrating how ML methods can enhance water quality assessment, strengthen pollution mitigation strategies, and promote sustainable river basin management. Future work will focus on integrating satellite-based land use and meteorological data to improve spatial analysis and extending the modeling framework to other river systems, thereby improving generalizability and applicability.

DOI: https://doi.org/10.5281/zenodo.17067318

 

Published by: vikaspatanker

Authors: Divyansh Singh Nikhil

Abstract: Hydrological modeling plays a pivotal role in addressing contemporary challenges of water resource management, particularly in regions facing rapid urbanization, climate variability, and increasing anthropogenic pressures. Among the widely adopted modeling frameworks, the Soil and Water Assessment Tool (SWAT) and its ArcGIS-integrated version, ArcSWAT, stand out as versatile, semi-distributed, process-based tools designed for simulating the impacts of land use, climate, and management practices on watershed hydrology. ArcSWAT has been extensively applied across continents, from small agricultural watersheds to large river basins such as the Mississippi, Nile, and Ganga, providing insights into surface runoff, evapotranspiration, groundwater flow, sediment transport, and water quality dynamics. Its integration with Geographic Information Systems (GIS) enables seamless spatial analysis, making it particularly suited to data-scarce basins in developing regions. This review synthesizes the development, structure, and functionality of the ArcSWAT model, with particular emphasis on its global and Indian applications. The analysis highlights key advances in calibration and validation approaches, including the use of SWAT-CUP and the SUFI-2 algorithm, as well as emerging practices of multi-site calibration to enhance model robustness. Special attention is given to case studies from India, including the Gomti River basin, where urbanization, agricultural intensification, and climate variability necessitate advanced modeling frameworks for sustainable management. The review identifies major advantages of ArcSWAT, such as its ability to handle large heterogeneous basins, perform scenario-based analyses, and integrate global datasets (e.g., SRTM DEM, Landsat LULC, FAO soils). However, limitations are also recognized, including data dependency, underrepresentation of water quality processes in many studies, and insufficient scenario-based applications in Indian contexts. Future research directions are outlined, focusing on coupling ArcSWAT with machine learning approaches, integrating climate change projections, enhancing parameter sensitivity and uncertainty analysis, and expanding hydrology–water quality modeling. By critically assessing past applications and current research gaps, this review establishes ArcSWAT as both a proven tool and an evolving framework for hydrological research. Its continued development and integration with emerging technologies hold the potential to transform watershed management and policy-making in the era of climate change and increasing water stress.

DOI: https://doi.org/10.5281/zenodo.17066923

 

Published by: vikaspatanker

Authors: Pankaj kumar singh, Professor Amit Thakur

Abstract: Energy utilization via every node is a significant concern in Wireless Sensor Network (WSN). Therefore, the main complexity deception in communicating the data that have the route with to the lowest degree distance as well as concentrates energy. Many investigators have residential different routing approaches for Cluster Head (CH) collection to communicate the packets to the BS. The choice of suitable CH, through the location also energy, is a main dispute in WSN. But, it can’t focuses on the network delay. Thus it decreases the network efficiency. To overcome this problem this paper Energy and data Communication delay aware Routing in WSN. Here, the fitness function is introduced for enhancing both the energy efficiency as well as lifespan of nodes through choosing the CH optimally. In this strategy, distance, energy, and delay of sensor nodes fitness function is used for selecting the optimal CH in the network. The network function is enhanced in this approach when equated to the conventional protocol.

Published by: vikaspatanker

Authors: Harsha Sammang, Harshini Balaga, Aditya Jagatha

Abstract: Coccidioidomycosis (Valley fever), caused by Coccidioides spp., is a climate- and soil-mediated respiratory disease whose exposure arises from inhalation of spores entrained by wind from disturbed, desiccated soils. Incidence is rising across the U.S. Southwest and expanding arid zones. Traditional surveillance is retrospective and weakly coupled to dust-generating processes (drought, grading, off-road activity), limiting actionable lead time for clinicians, public health, and occupational safety. We present Aeolus-DS, a Design Science Research (DSR) artifact that fuses aerosol remote sensing (MAIAC AOD; dust fraction), mesoscale meteorology and soil moisture (ERA5), land-disturbance telemetry (construction and energy activity; off-highway vehicle events; nightlights), and clinical sentinel signals (syndromic ED chief complaints; pneumonia rule-out) into a dust-aware, AI-driven early warning and deci- sion support system. Methodologically, we propose a graph spatiotemporal transformer with direction-aware attention and physics-guided regularization reflecting aeolian transport. Us- ing county–week panels (2014–2024) for AZ–CA–NV, Aeolus-DS improves nowcasting MAE by 18% and two-week AUPRC by 21% over strong baselines (XGBoost, LSTM). Role-based “action cards” translate probabilistic forecasts and uncertainty into targeted mitigations (site watering cadence, temporary grading pauses, N95 staging, clinician test prompts). We eval- uate predictive skill, calibration, runtime, interpretability, and stakeholder usability, and discuss governance, ethics, and portability to other dust-borne mycoses in climate-stressed regions.

DOI: https://doi.org/10.5281/zenodo.17063019

Published by: vikaspatanker