VFAST - Virtual Foundation for Advancement of Science and Technology (Pakistan)
Not a member yet
1255 research outputs found
Sort by
A Novel Approach toward Windspeed Forecasting using an Advanced Deep Learning Framework with Explainable AI
Full text linkAccurate wind speed forecasting is vital for optimizing renewable energy deployment and for advancing our understanding of climate dynamics. Traditional machine-learning approaches often neglect the fundamental physical principles driving atmospheric processes, which limit their robustness and ability to extrapolate beyond the training domain. This investigation introduces an innovative PINN architecture that integrates deep learning techniques with established meteorological theory to improve both predictive fidelity and interpretative clarity. The framework embeds a temperature-sensitive physical constraint directly within the optimization objective. This formulation guarantees that the predictions remain consistent with thermodynamic equilibrium. Structured as a four-layer sequential network with 13 inputs, two hidden (64 neurons each), and a single output unit, the PINN outperformed eight competitive baseline architectures ranging from Bayesian ridge regression to gradient boosting and various hybrid architectures trained on a suite of handcrafted covariates, including wind-shear terms, mean humidity, and temperature-interaction derivatives derived from multi-year records spanning the climatically distinct locales of Badin, Dadu, and Rohri. Observed reductions in mean-squared error and mean absolute error were dramatic, and the coefficient of determination rose to an impressive 0.99. Furthermore, the application of XAI techniques, specifically SHAP and LIME, identified temperature and humidity as the dominant predictors, corroborating the physical consistency of the model while ensuring operational transparency for users. This study establishes an integrative linkage between data-driven learning methodologies and established domain expertise, resulting in a robust and interpretable decision-support tool for both energy system planning and climate impact assessment
Solutions of Real-World Problems Using Transportation Method
Full text linkThe paper is an analysis of the transportation problem which is one of the most basic and popular optimization problems in operational research. The transportation problem is aimed at finding the most cost-efficient means of distributing a homogenous product, usually a factory or origin, to various demand points, usually a warehouse or destination. Every source is characterized by the fixed amount of supply, and every destination is characterized by the particular demand that needs to be addressed. The main aim of the transportation model is to reduce the overall transportation cost and at the same time make sure that all the supply and demand restrictions are met. It is assumed that the means of transporting a unit of product to a specific warehouse is known and constant. The decision variables are the amount of goods carried within each route generated by variations of the origins and destinations. Through the proper choice of these variables, the model determines the most optimum plan of transportation that will give the lowest overall cost. This paper is the formulation of the mathematical model of the transportation problem, its objective function, and the constraints associated with it. The objective function can be established as the result of multiplication of the unit transportation costs and the decision variables of all routes. The constraints are designed to make sure that the overall number of goods shipped by each factory will not be beyond what it can supply and also the overall amount that is received by each warehouse will satisfy its needs. There are also non-negativity restrictions that are applied to ensure that all amounts of shipments are practical. To understand it better, the simplistic model of the transportation issue is depicted through the tabular representation, in which the costs, supplies, demands, and allocation variables are evident. The constraints and decision variables interaction is discussed to show how feasible and optimality solutions are obtained. On the whole, this paper argues that the transportation issue is significant because it is a valuable and effective instrument of streamlining the distribution systems in the practical logistic and supply chain tasks
Determinants and Forecasting of Islamic and conventional Banks Profitability in Pakistan by using Logistic Regression
Full text link Banks’ profitability has been an essential consideration for economists, investors, and researchers. It serves as a benchmark for the banking industry economy. In the previous literature, it was observed that an internal factor of Pakistan banks’ profitability i.e. non-interest income was not included. This study examined in addition to other micro and macro factors, the non-interest income of banks’ profitability in Pakistan. Data for the period 2007 to 2016 was obtained from State bank Pakistan and the Finance Division of Pakistan. Bank’s profitability was measured by the return on assets taken as the dependent variable. An unusual technique for such studies, logistic regression was applied. Although, by default return on asset is a continuous variable it was converted to categorical by assigning “1” for profitable banks and “0” for the non-profitable bank. Results indicated that bank size, operating efficiency, and interest ratio have significant effects on the profitability of banks. Interest ratio and operating efficiency have a negative relationship with return on asset and the size has a positive effect on return on asset. It was found that non-interest income has no significant effect on banks’ profitability
Advancement in Smart Vision Systems: A Computer Vision-Based Assistive System for Visually Impaired Individuals
Full text linkVisually impaired individuals experience significant mobility challenges due to the limited situational awareness provided by traditional aids like white canes. To address this, an AI-powered smart vision kit that enhances environmental perception through real-time object detection and audio feedback has been proposed. Our system combines embedded edge computing with optimized neural networks to deliver a portable, low-cost assistive solution. The hardware prototype incorporates a Raspberry Pi 4 and camera module with a TensorFlow Lite pipeline, utilizing a quantized MobileNetV1 SSD model trained on the COCO dataset for efficient inference. The framework processes live video streams via OpenCV, detecting objects within a 5-meter range at 12 FPS (tested on 480p input). Detections are converted to spatialized audio alerts using text-to-speech (TTS), prioritizing critical obstacles.
Eclipse Application Programming Interfaces: How Buggy Are They?
Full text linkEclipse Framework provides stable public APIs and unstable internal APIs. However, there is no guarantee that these interfaces are well tested because several bugs are reported by interface users on Bugzilla-based Eclipse project. Applications that use buggy APIs risk failing if bugs are not fixed. Bug fixation and resolution takes at least 3 years thus API users have to fix the bugs themselves or abandon that particular API. The study aimed at identifying bug free interfaces in the Eclipse Framework and recommend them to application developers. In this research study, we used both SonarQube and SpotBugs static analysis tools to carry out an empirical investigation on 28 major Eclipse releases to establish the existence of bug free interfaces. We provide a dataset of 218K and 303K bug-free public API and internal API respectively. There exist over and bug-free public APIs and internal APIs, respectively, in Eclipse releases. Furthermore, over 80.8% and 44.2% are major and Malicious code vulnerability bugs respectively and the average bug remediation effort is 105 days. Results from this study can be used by both interface providers and users as a starting point to know tested interfaces and also estimate efforts needed to fix bugs and an online dataset of bug-free interface is available on Github for developer
Securing Web Applications: A Practical Approach to Mitigating OWASP Top 10 Vulnerabilities
Full text linkThe exponential growth of online platforms and application has made us believe that securing web apps is more important to mitigate attacks viz., data breaches, frauds, unauthorized access etc. But web applications are still vulnerable in a number of ways that can be abused by attackers. In this context, we propose a pipeline to identify and reduce security threats in web applications, focusing on the OWASP Top 10 vulnerabilities — highly publicized risks with clear exploitation vector; namely: injection attacks, broken authentication, sensitive data exposure or cross-site scripting. For every vulnerability, we cover them with practical demonstrations by using BeeWAP (Beehive\u27s Educational Web Application Platform), an intentionally vulnerable web application for the Web testing and security education purpose. The vulnerabilities are analyzed based on real-world contexts in BeeWAP platform, which helps to assess the implications of web application security. We are using techniques of standard tools like Burp Suite to find these weak points and also implementing countermeasures, hence gives an all-in-one manual focused on securing applications from threats.It elaborates a methodology to identify vulnerabilities, perform risk analysis to develop security models that respond specifically to the identified OWASP Top 10 vulnerabilities. In this paper, we demonstrate real-time risk mitigation by simulating common attack vectors and showing the resulting insight into good practices for securing web applications. In this direction, the present paper tries to step forward towards reconciliation between theory and practice, by providing a structured model that represents a compromise that security personnel and developers can use directly in order to improve defensive capability in applications. More specifically, our results emphasize the importance of constant vulnerability testing and continual training of cybersafety measures on protected infrastructures. These practices, when enacted by developers, can bolster defenses against the ever-evolving nature of cyber threats and ultimately lead to more trustworthy and reliable web applications
Advances in Multilevel Encryption Techniques: A Comprehensive Review of Hyperchaotic Neural Networks, Quantum-Inspired Approaches, and Data Hiding Mechanisms
Full text linkThe growing sophistication of cyber threats and the limitations of traditional cryptographic methods have necessitated the development of advanced encryption frameworks. This paper presents a comprehensive review of multilevel encryption techniques, focusing on three key dimensions: hyperchaotic neural networks, quantum-inspired encryption (QIE), and advanced data hiding mechanisms. Hyperchaotic neural networks, characterized by their high-dimensional chaotic systems and dynamic adaptability, generate unpredictable key sequences to enhance resistance against brute-force and statistical attacks. Quantum-inspired encryption leverages principles such as superposition and entanglement to design lightweight, scalable cryptographic frameworks that operate on classical systems, offering high entropy and robust security for IoT and real-time applications. Additionally, adaptive data hiding techniques, including neural network-based steganography and hyperchaotic embedding, ensure imperceptibility and resilience against compression and detection. This review consolidates state-of-the-art advancements, comparing the performance, scalability, and application of these techniques across domains such as healthcare, IoT security, multimedia protection, and cloud storage. The integration of these approaches into multilevel frameworks is highlighted, along with their potential to address computational, scalability, and security challenges posed by modern cyber threats. Future research directions are identified, emphasizing the development of hybrid techniques, energy-efficient algorithms, and robust implementations for emerging applications in cybersecurity and beyond
New Numerical Insights into Electromagnetic Mass in Spherically Symmetric Configurations
Full text linkOur study aims to evolve artificial intelligence that emerges from natural processes and be able to predict nonlinearities in the lane Emden-Fowler (LEF) equation. More precisely, the feedforward artificial neural network model is an adaptive one that leads to accurate solutions of the LEF equation. This dataset concerns a neural network whose parameters have been made adjustable so that initial predictions can be based on a given model quite easily. The energy reduction objective function for the specific limitations of the LEF equations is based upon contextual nuances introduced in previous optimization process. To begin with, this proposed methodology has been tested through experiments and initial conditions affecting the initial value of any such problem for LEF. We consider three cases in detail which show how our method can solve the LEF equation effectively. Our combined method (PSO-GWO-IPA) with Particle Swarm Optimizer (PSO) and Grey Wolf Optimizer (GWO) achieves very good convergence speed when compared to PS, IPA, PSO, PS-IPA, HPM and OHM. Through statistical tests, we will verify reliability and validity of our approach mentioned above. Our empirical results are in perfect agreement with the mathematical model, demonstrating the wisdom of the proposed method
Investigating Reactive Species Diffusion in Power-Law Fluids Beyond Stretched Surfaces
Full text linkIn this study, a numerical solution is developed for the steady flow of a power-law fluid influenced by magnetohydrodynamics (MHD) effects over a continuously moving, linearly stretching surface. The model incorporates mass transfer involving species concentration and a one-stage chemical reaction under isothermal and homogeneous conditions. Reactive species emitted from the surface diffuse into the fluid, driven solely by the sheet\u27s motion.The model examines the effects of six key parameters: power-law index, magnetic parameter, modified Schmidt number, reaction rate parameter, wall concentration parameter, and local Grashof number. These factors significantly influence the mass transfer characteristics and fluid flow behavior.To solve the governing equations, an artificial neural network (ANN) combined with the Levenberg-Marquardt Algorithm (LMA) is employed. The differential equation is reformulated into an ANN-based optimization problem, minimizing the mean square error as a fitness function. The ANN-LMA results are compared graphically and numerically with reference solutions for validation.Graphical results show that the magnetic field increases surface skin friction but slightly decreases the mass transfer rate. However, the mass transfer is highly sensitive to changes in the modified Schmidt number and the reaction rate parameter, increasing with their higher values. Statistical measures such as mean, standard deviation, minimum, and maximum values are used to further validate the solution.Overall, the study highlights the importance of exploring non-Newtonian fluid models, especially those exhibiting shear-thinning behavior, which appears to lower wall shear stress and adds complexity to power-law fluid dynamics
Urdu-Punjabi Code Switched Sentiment Analysis Empowered by a Deep Learning Framework Integrating XLM-R, and GPT
Full text linkSentiment analysis is a procedure that uses computational methods, textual analysis, and natural language processing to derive significant insights from textual sources. Sentiment analysis detects and quantifies the attitudes, opinions, and emotional states that individuals convey through textual information. The majority of existing sentiment analysis work is centered on the English language, leaving low-resource languages largely underexplored. Performing sentiment analysis on low-resource languages is challenging due to the unavailability of extensive datasets and associated resources. To overcome the challenge of unavailability of datasets we proposed Large Urdu-Punjabi code switched Corpus for Sentiment Analysis (LUPCSA-25) comprises over 10,00,000 user reviews in Urdu and Punjabi (Shahmukhi). Urdu and Punjabi domain specialists enrolled in PhD provided additional annotations to the dataset. In this research, we examine how head-pruning strategies can enhance both the predictive accuracy and computational efficiency of transformer architectures—specifically XLM-R and GPT-2—for sentiment classification of Urdu–Punjabi code-switched text. After preprocessing the textual data, BERT embeddings are produced and subsequently passed to the proposed classification model for determining sentiment. The performance of the proposed classifier is assessed by comparing it with baseline classifiers. The results demonstrate that the proposed classifiers with head pruning technique surpass current state-of-the art models with a precision rate of 96.4%