
Biomonitoring water quality is an essential method for maintaining the ecological health of resources. This study aims to analyze the temporal trends over a period (1984-2022) of four water quality-related indices (LST, TSS, NDCI, NDTI) in three Algerian dams (Beni slimene, Ledrat, Ghrib) using space remote sensing via the Google Earth Engine platform. The dataset is subjected to Mann-Kendall statistical tests and linear regression to detect any significant trends in the indices studied. The results show a clear opposition between the two-time scales. On a seasonal scale, there is no statistically significant trend in the three dams (p-values > 0.05; R² close to zero), indicating the absence of systematic seasonal cyclicity. On the other hand, the annual analysis reveals marked and highly significant trends: surface temperature (LST) and suspended solids (TSS) increase in all three dams (p < 10⁻⁵ for LST; p < 10⁻⁹ for TSS). The NDCI index (chlorophyll/organic matter) increased significantly in Beni Slimane (p = 3.9×10⁻³) and Ledrat (p = 1.24×10⁻⁵), but remained stable in Ghrib. The NDTI index (turbidity) decreases only at Ghrib (p = 5.52×10⁻⁵), with no detectable trend in the other two dams.
This paper deals with a system of fractional stochastic differential equations with delay. Under some regularity conditions we study the well posedness of the system and we establish the existence of stochastic optimal control.
This studyintroduces a refined approach to farming with sunlight, combining solar panels and crop growth while using connected sensors through long-range wireless signals alongside wiring data transfer. Farming under solar arrays saves space by producing food and electricity together on shared ground. Still, current setups struggle due touneven water delivery, unstable power handling, weak links among devices, and outdated plant behavior predictions when weather shifts occur. One wayaround these drawbacks begins with anewdesign:an IoT-based smartwater control setup paired with dynamic crop models. Instead of relying on extra wiring, power line communication moves data quickly using current electrical lines. For field sensors spread far apart, LoRaWAN handles connections efficiently - long reach, little power needed. Data flows nonstop - from earth dampness toweather shifts to plantstate - shapingsmarter watering plans. Health tracking for crops gets sharper. Energy use adjusts in response. Scaling up stays affordable. A key part of the system lies in its ability to fine-tune fertilizers according to both soil conditions and how crops develop over time. Because nutrients are applied more precisely, yields rise -alongside better use of water - while excess runoff and ecological harm drop off. Testing via MATLAB models, together with real-world examples, confirms gains across farming output, smarter watering practices, and improved energyoperations. When combined, programmable logic controls, long-range wireless communication, and smart nutrient handling open doors - not just for growing food sustainably - but also for sharing land wisely between farming and clean power generation.
Aim: This study aims to explore how economic, environmental, and social sustainability efforts influence the performance of a supply chain and to investigate how the role of government regulations and SDG 12 (Responsible Consumption and Production) may impact these effects. Method/Design: The study employed quantitative methods and utilised data collected from 365 professionals working in supply chains across various fields. The researchers used a questionnaire to assess factors such as sustainability dimensions, the overall regulatory framework, the incorporation of SDG 12, and performance in the supply chain. To assess the proposed hypotheses, structural equation modelling (SEM) was conducted using SmartPLS. Findings: The results indicate that environmental and economic sustainability practices have a positive influence on supply chain performance, whereas social sustainability does not show a significant impact. The rules and regulations in place had a direct effect and played a significant role in shaping how economic sustainability affected the company's performance. But none of the sustainability-performance relationships were significantly affected by SDG 12. They emphasise that strong institutions and economic cooperation are crucial for achieving better results in sustainable supply chains. Conclusion: This study brings a unique perspective to the field by examining sustainability in global supply chains through the lens of multiple theories, incorporating regulations and SDG 12 into the analysis. The research results help companies that aim to compete by following specific sustainability plans and understand how to interact with regulators.
Introduction: The process of cloud infrastructure management remains a high-friction and cognitively intensive endeavor since the Infrastructure-as-Code (IaC) platforms are platform-specific, syntactically dense, and coupled with provider-specific configurations. Since organizations are rapidly moving to cloud-native architecture to ensure they provide scalable and distributed applications, the operational cost of configuring, validating, and maintaining infrastructure has increased manifold. Manual provisioning processes demand extensive technical knowledge and they tend to have repetitive documentation reading, which makes configuration drift, deployment bugs, and ineffective operations more probable. These obstacles pose both a hindrance to enterprise-level efforts at implementing DevOps as well as to students and practitioners in need of a public, low-risk environment in which to learn and test out cloud technologies. Objectives: This paper seeks to create an AI-agent-based architecture, Cloudent, that provides a translation between natural-language user intent and executable cloud infrastructure by automating the process of IaC generation, enhancing deployment reliability, and cutting operational overhead by generating Intelligent Reasoning and Self-Correcting. Methods: Cloudent brings together an Agentic AI reasoning engine, based on LangChain and LangGraph, with the Pulumi Automation SDK within a Next.js application to write programs generating type-safe TypeScript IaC without any external CLI. The semantic retrieval layer translates the will of the user into cloud provisioning profiles and an iterative self-healing process translates the deployment logs and feedback on errors and adjusts configurations autonomously. The emulation based on LocalStack provides safe and cost-transparent environment of testing. Results: Analysis of the suggested framework indicates a high level in automating processes of infrastructure provisioning and still `maintaining contextual accuracy and operational consistency. The combined model recorded training accuracy of 96.50% and validation accuracy of 94.50% in interpreting and executing infrastructure tasks meaning that it was reliable in translating natural-language requirements into deployable resources. The self-healing system minimized the number of debugging cycles, increased the effectiveness of the deployment process, and allowed manual corrections to be performed iteratively, which speeded up the provisioning process and increased trust in automated DevOps operations. Conclusions: Cloudent illustrates how Agentic AI can change the DevOps processes through transforming the conversational requirements into production-deployable infrastructure. The structure provides a democratized and scalable cloud management solution through autonomous reasoning, correction through iteration and safe emulation, which increases the rate of deployment and improves confidence in operations.
Artificial intelligence is playing vital role in various domains presently. Multiclass classification is an important task of AI. The concept of multiclass classification has been adopted in various domains such as medical science, banking sector, corporate, cyber security etc. Still researchers are applying such concept in different fields and trying to enhance the accuracy and efficiency of the models. The efficiency and accuracy of models are depend on various factors of such as dataset, selection of features, selection of algorithms, selection of hyper-parameters. The main objective of this paper is to present the various machine learning algorithms that are used for multiclass classification besides their merits and demerits. The present study also states the various research gaps in the existing multiclass classification models that are need to be resolved. This study will be useful to the researchers who are applying the multiclass classification in a specific domain to classify the data with great accuracy and efficiently. [ The efficacy of models used for classification, recognition, diagnosis, or clustering of data of different domains is determined by their performance in real-world applications. Evaluating such models require detail understanding of their underlying equations and features to discern whether they perform "well" or "poorly." Such methods have become increasingly important to researchers in recent years. While a wide range of statistical methods has been applied, there remains a gap in guiding researchers toward the most appropriate method for their specific applications. This paper collects and analyzes the most significant classification methods used in research. It provides a comparative analysis of these methods, focusing on their mathematical foundations, purposes, and limitations. The study highlights that feature selection plays a significant role in classification performance, and the results provide guidance on selecting features for classification, recognition, diagnosis, or clustering tasks.]
This paper aims to study the complex nonlinear dynamics involved in the financial system, depending on three key parameters: saving amount, the cost associated with unit investment and elasticity of commodity demand. This research explores the formation of a chaotic attractor by altering a parameter within the financial model. We illustrated that a Hopf bifurcation takes place, resulting in the emergence of a stable limit cycle. Previous studies have not addressed the quasi periodicity and period halving bifurcation. Through numerical analysis, we have uncovered a cascade of period halving bifurcation, quasi periodicity which led to the formation of a strange attractor. The existence of chaos in the system has been effectively recognized using various methods, including bifurcation diagrams, lyapunov Dimension, lyapunov Exponents, time series and, two dimensional and three-dimensional phase portraits. The system's sensitivity is estimated utilizing the fourth order Runge-Kutta method.
This study investigates the periodic and chaotic dynamics of an amplitude-modulated (AM) memristive circuit using numerical simulations based on the fourth-order Runge-Kutta method. The system’s response is analyzed under varying modulation parameters, revealing a rich spectrum of nonlinear phenomena, including period-doubling and reverse period-doubling bifurcations, period-bubbling, quasiperiodic oscillations, and chaotic behavior. Bifurcation diagrams, phase portraits, and Poincar’e sections confirm the presence of chaotic attractors and complex dynamical regimes. The frequency and amplitude of the modulating signal play a pivotal role in shaping the system’s behavior, influencing both the onset of chaos and the structure of the resulting attractors. Furthermore, the memristive nature of the circuit is verified through the presence of pinched hysteresis loops in the voltage-current (v - i) characteristics-an essential signature of memristive systems. These loops exhibit distinct frequency- and amplitude-dependent variations, underscoring the nonlinear, memory-dependent behavior inherent to the memristor.
In digital electronics Code converters are logic circuits that translate data from one binary code format to another (e.g., Binary to BCD, BCD to Gray, BCD to Excess-3) to enable communication between systems. The paper n-bit Code Converters Using Verilog includes three Decoders 4-bit Gray to Binary and 4-bit Binary to Gray Code Converters. The two Code Converters are designed, simulated and Synthesized Using Verilog. The Verilog Modules of each Code Converter are developed and Synthesized to obtain RTL and Technology schematics. In the next step, The Verilog Test benches are developed for each Code Converter and simulated using Behavioral simulation to obtain the Output waveforms. Next The Output waveforms are verified as per the given Truth Tables. The design summary of each Code Converter can be obtained after synthesization and simulation. The design summary includes Timing Summary, Device Utilization Summary, Primitive and Black Box Usage and Timing Reports etc. In future n-bit Code Converters can be further implemented for increased value of n with all possible input combinations. n-bit Code Converters can be designed Using VHDL as well as other HDL languages also and the design can be implemented using Field Programmable Gate Array(FPGA).
In this work, we explore the stability of weak solutions to a stochastic version of a globally modified coupled Cahn-Hilliard-Magnetohydrodynamic model with multiplicative noise. The model describes the flow of the mixture of two incompressible, immiscible fluids under the influence of an electomagnetic field with stochastic pertubations. This system consists of the globally modified Magnetohydrodynamic model for the velocity and magnetic field, coupled with a Cahn-Hilliard equation for the order (phase) parameter. We prove that the weak solutions converge exponentially in the mean square and almost surely exponentially to the stationary solutions.We also show a result related to the stabilization of these equations.
Introduction: Emotion detection from text is a fine-grained natural-language-processing (NLP) task with growing relevance in affective computing, mental-health informatics and digital customer-experience analytics. Despite significant progress in English, fine-grained emotion recognition for Hindi — a morphologically rich, medium-resource language — remains comparatively underdeveloped and most published "cross-lingual" work actually reports parallel monolingual results without measuring transfer. This paper contributes a reproducible bilingual emotion-detection benchmark built entirely on publicly available, peer-reviewed datasets: GoEmotions for English and BHAAV supplemented with XED-Hindi for Hindi. Both source datasets are mapped to a harmonised five-class Ekman-aligned emotion scheme {joy, sadness, anger, fear, surprise}, with a transparent label-mapping protocol released for reproducibility. We propose a Transformer–Recurrent ensemble that fuses a multilingual Transformer encoder with a BiLSTM–GRU branch operating on pretrained FastText/IndicFT embeddings through a weighted log-probability fusion, systematically benchmarking four multilingual encoders (mBERT, XLM-R, MuRIL, IndicBERT). We further conduct a genuine cross-lingual evaluation in which models trained on English GoEmotions are tested zero-shot and few-shot on Hindi BHAAV, quantifying transfer rather than assuming it. On the harmonised supervised task, the MuRIL-based ensemble attains a macro F1 of 0.72 on English GoEmotions and 0.58 on Hindi BHAAV + XED, matching standalone MuRIL on both languages and remaining within one percentage point of the strongest single-model baseline (XLM-R, 0.73 on GoEmotions). MuRIL emerges as the strongest single encoder on Hindi (0.58 vs 0.54 XLM-R, 0.53 mBERT, 0.42 IndicBERT), confirming the value of Indic-specific pretraining. In the zero-shot English→Hindi setting, the ensemble recovers a macro F1 of 0.19, rising to 0.50 with 20% of BHAAV labels and 0.58 at full supervision; the smooth few-shot improvement curve (0.38, 0.45, 0.50, 0.54 at 5%, 10%, 20%, 50% Hindi data) confirms that cross-lingual transfer is genuinely useful once even a small amount of target-language supervision is available. Ablations, confusion-matrix analysis, and error analysis identify anger and joy on Hindi as the hardest classes, with the heavy fear/non-fear class imbalance in BHAAV (8,188 vs ≤2,200 samples per class) and the cross-domain mismatch between Reddit social media (English) and Hindi literary prose accounting for the largest share of errors.
This work presents a hybrid federated learning framework for stress detection using physiological signals collected from wearable sensor devices. The system utilizes the WESAD dataset, which includes multimodal physiological signals such as electrocardiogram (ECG), electrodermal activity (EDA), respiration, and body temperature. A binary classification task is performed to distinguish stress and non- stress conditions after signal preprocessing, normalization, and feature preparation. Feature selection is carried out using a Random Forest classifier to identify the most relevant physiological attributes and reduce dimensionality. A centralized Multilayer Perceptron (MLP) model is first developed as a baseline, followed by a Federated Learning (FL) framework in which multiple client models are trained locally without sharing raw data. To address heterogeneity in client data distributions, Clustered Federated Learning (CFL) is introduced by grouping clients based on model similarity before aggregation. In addition, a Quantum Federated Learning (QFL) approach is incorporated to further enhance the learning framework for stress classification in privacy-preserving settings. Experimental evaluation is conducted under centralized, federated, clustered, and quantum federated settings, and performance is measured using accuracy, precision, recall, and F1-score. The framework demonstrates effective stress detection performance while preserving privacy by ensuring that sensitive physiological data remains decentralized.
This work is devoted to the study of multidimensional backward doubly stochastic differential equations under weak regularity assumptions on the coefficients. In particular, we introduce a new class of non-Lipschitz conditions and show that the associated BDSDEs admit a unique solution in Lp for any p > 1. Several classical results are recovered as particular cases of our analysis.
The Zika virus (ZIKV) poses significant public health challenges due to its rapid transmission and severe effects on pregnant women and new-borns. In this paper, we develop a compartment mathematical model to study the transmission dynamics of the Zika virus between human and mosquito populations. The human population is divided into susceptible, exposed, infected, pregnant, infected infant, and recovered classes, while the mosquito population includes susceptible, exposed, and infected compartments. The basic reproduction number is derived using the next generation matrix method to determine the threshold condition for disease persistence. By examining the existence and uniqueness, local and global stability analyses of both the disease free and endemic equilibrium are performed using linearization and Lyapunov function techniques. We incorporated optimal control to identify effective prevention, treatment, and vector management strategies in the model. To obtain approximate analytical solutions, we employ the q − homotopy analysis transform method (q − HATM), which provides a rapidly convergent series solution without restrictive assumptions. The results demonstrate that reducing the contact rate and enhancing recovery significantly decrease and prevent disease outbreak.
The system develops an iot based obstacle detection and warning system which improves safety and awareness while enabling visually impaired people to move freely .The system operates through continuous environmental surveillance which uses various sensors that includes distance sensors and echo sensors and passive infrared motion sensors and flame detection .An ensemble machine learning system with voting classifier assess real time sensor data to identify four different environmental states which includes normal and warming and alarm and voice and collision.The system uses a safety override mechanism based on rules to handle high -risk situations which creates an alert when an obstacle comes with in 10 cm of the machine learning system . The system performs data preprocessing and feature scaling before classification to enhance prediction accuracy .The system delivers real time multi-channel alert through asynchronous audio notification that use text to speech technology to provide clear voice feedback about obstacle proximity and motion detection and fire hazard .A remote monitoring module send real time alert which include sensor data and threat assessment to a communication platform for ongoing observation and incident documentation the system features a complete training and evaluation system which tests various classifiers together with the final ensemble model while it automatically creates performance assessment results . The experimental results establishing that the system successfully identifies threats and send alerts on schedule which proves its effectiveness as an intelligent assistive system for users who are visually impaired
Attention Deficit Hyperactivity Disorder (ADHD) is a common neurodevelopmental condition marked by attention deficits and impulsivity. Conventional pharmacological and behavioral treatments often lack personalization and could cause side effects. Low-intensity focused ultrasound offers a non-invasive neuromodulation alternative, though complex ultrasonic wave propagation in brain tissues limits precise targeting. This research presents a Golden Jackal Optimized Deep Spiking Neural Network (GJO-DSNN) framework for intelligent prediction of ultrasonic wave dynamics in ADHD neurotherapy. The DSNN models temporal neuro-acoustic interactions using biologically inspired spike-based learning, while the GJO adaptively tunes network parameters to enhance prediction stability and focal accuracy. The framework utilizes multimodal neuroimaging data, simulated ultrasonic propagation profiles, and neurophysiological recordings acquired across multiple therapy sessions. Data preprocessing is performed using Z-score normalization to ensure uniform signal scaling. Discriminative spatiotemporal features are extracted using the Discrete Wavelet Transform to effectively represent acoustic and neural response characteristics. Performance evaluation based on prediction accuracy (97.5%)and neurophysiological response consistency demonstrates reliable and adaptive wave-focusing behavior. The framework is implemented using Python-based neural modeling and signal processing tools. The proposed approach establishes a personalized, adaptive, and non-invasive ultrasonic neurotherapy paradigm for ADHD management with strong potential for clinical translation.
The primary objective of this paper is to examine the -symmetries of indefinite almost paracontact metric manifolds, with particular emphasis on three-dimensional -para Kenmotsu manifolds. We demonstrate that the Einstein manifolds and the stipulation of possessing distinct -symmetries are equivalent for three-dimensional -para Kenmotsu manifolds. It has also been deduced that a -pseudosymmetric three-dimensional -paraKenmotsu manifold exists. In the conclusion, we prove that all projectively -semisymmetric three-dimensional -para Kenmotsu manifolds are Einstein manifolds.
Mobile ad hoc networks (MANETs) have attracted increasing attention in the field of computer science due to their many civil and military applications. However, these applications pose significant security challenges. To effectively protect mobile wireless networks, it is crucial to implement an intrusion detection system (IDS) that is tailored to the specific characteristics of these networks and capable of identifying various types of threats. This article proposes an approach for designing an intrusion detection system, in which a dataset specifically designed for mobile ad hoc networks (MANETs) is generated to improve black hole attack detection. This approach exploits the functionalities of different layers of the OSI model, such as the physical, MAC, and network layers, to extract relevant features. A scheme was developed to collect data from the Network Simulator 2 (NS-2) simulator and process it to form a dataset called MANET-DataSet. An artificial neural network (ANN) was then trained with this dataset to detect black hole attacks. The results show that using MANET-DataSet improved the accuracy of the intrusion detection system while reducing the false positive rate.
With the use of a bipolar fuzzy graph, we examined the concepts of vertex and edge cardinality in complex bipolar fuzzy graphs. We also defined dominating set, independent set, and total dominating set in complex bipolar fuzzy graphs and looked at minimal domination and maximal independence. Additionally, talk about the characteristics of the neighborhood and its dominance.
Image denoising is an essential preprocessing task in many computer vision and imaging applications. The existing spatial and transformation-based filtering approaches often fail to be generalizable across different noise models due to their homogeneous filtering approach applied over the entire image. In this paper, we propose a new image denoising method using a framework formulated based on Markov Decision Processes (MDP), solved through Tabular Q-learning algorithm. We consider individual pixels to be reinforcement learning agents interacting with its local environment and choosing the best action from the action space consisting of various filter types (Median, Gaussian, Bi-lateral, Wavelet Filtering). As a solution to the issue of high-dimensional continuous states (pixels' neighborhoods), we design a State Quantization process utilizing local statistics (mean, variance) of individual pixels. As a result, our RL agent learns to find the optimal solution using the Tabular Q-function. Empirical evaluation shows that compared to traditional single filtering, our pixel-wise adaptive filtering technique achieves better PSNR and SSIM values with edge preservation. Also, this research paves the way for further development of DRL methods for continuous state space problems.