VFAST - Virtual Foundation for Advancement of Science and Technology (Pakistan)
Not a member yet
1255 research outputs found
Sort by
An Efficient System for Urdu Sign Language Recognition using Support Vector Machine(SVM), Convolutional Neural Network (CNN), and Ensemble Machine Learning (EML)
Full text linkSign language has significant problems in the everyday life of deaf and hard-of-hearing people. We have used a Support Vector Machine (SVM), a Convolutional Neural Network (CNN), and an Ensemble Machine Learning (EML) model that combines their outputs as our machine-learning technique. We seek to design a USL recognition system that will address communication gaps. Firstly, we reviewed the “Dataset of Pakistan Sign Language and Automatic Recognition of Hand Configuration of Urdu Alphabet, through Machine Learning”. The dataset has various characteristics, such as image quality, size, and class distribution. The dataset plays a pivotal role in training and evaluating the proposed models. It includes a diverse range of images representing the Urdu Sign Language (USL) alphabet, ensuring the models are exposed to varying hand configurations, backgrounds, and lighting conditions. This diversity helps improve the generalizability of the trained system. During preprocessing, we performed normalization, resizing, and augmentation techniques to enhance the robustness of the data and prevent overfitting. Results indicated that the ensemble approach outperformed the individual models, achieving higher classification rates for several challenging hand configurations. The developed system shows promising potential for real-world applications in bridging the communication gap faced by the deaf and hard-of-hearing community in Pakistan
Leveraging Software Engineering Principles: An IT-Enabled Component-Based Knowledge Management Model for Healthcare Information Systems in Developing Countries
Full text linkIn the contemporary era, the integration of Information Technology (IT) with Knowledge Management (KM) infrastructures has significantly transformed the healthcare sector. This study investigates the current use and effectiveness of IT in supporting KM functions within public-sector healthcare organizations in Pakistan. It further explores and evaluates key factors influencing the success of KM practices in this context. Employing a triangulated mixed-method approach, the research examines the maturity level of KM within healthcare information systems (HCIS) and the general understanding of KM concepts among stakeholders. Empirical findings reveal that most of the proposed hypotheses were statistically significant, except for two: the impact of exogenous factors on KM specifications and the impact of KM success on overall organizational performance, which were found to be weak and non-significant. A conceptual model was initially develop based on the hypothesized relationships, followed by an empirical model tested through Partial Least Squares Structural Equation Modeling (PLS-SEM) to assess significance (p < 0.05) and model fit. Based on these analyses, the study proposes an integrated, component-based KM model aimed at enhancing the governance and effectiveness of HCIS
Smart Water Quality Management System: A Case Study of Tharparkar Region
Full text linkIn this paper, we present the design and implementation of a smart water quality monitoring system for the Tharparkar region of Pakistan, where access to clean water is limited. The system utilizes Internet of Things (IoT) sensors and machine learning algorithms to assess and predict water quality. Parameters such as pH, turbidity, and total dissolved solids were continuously monitored using IoT sensors deployed in three strategically selected groundwater wells in Tharparkar. The collected data was transmitted wirelessly to a central server, where a Support Vector Regression model was applied to analyze water quality trends and classify samples as polluted or unpolluted. The results demonstrate the system\u27s effectiveness in providing accurate, timely, and location-specific information, enabling early detection of contamination, and supporting proactive water resource management. This work highlights the potential of integrating IoT and artificial intelligence to address water scarcity and quality challenges in an underdeveloped region
Enhancing Honey Quality Control: A Machine Learning-Based Approach Using Hyperspectral Imaging
Full text linkFood safety and quality control is an important component of sustainable healthcare in smart cities as it contributes to public health, preventive healthcare, sustainable agriculture, consumer empowerment, economic development, and regulatory compliance. The focus of this paper is on adulteration identification of honey. Honey is a natural sweetener that has been utilised for thousands of years for its different health advantage such as culinary sector, skincare, wound treatment, and as a natural cough suppressant. Adulteration of honey refers to the practice of introducing contaminants or diluting pure honey with other substances such as sugars, syrups, or water in order to increase volume and lower manufacturing costs. There are different mechanisms for identifying adulterated honey e.g physiocochemical properties, chromatography, spectroscopy, and hyperspectral imaging; each of which presents its own sets of challenges and limitation. The current study uses a publicly available dataset with different types of honey adulterated with sugar syrup. Hyperspectral imaging is used to extract spectral features of the honey samples. As the dataset represents an unbalanced representation of the adulterated samples. We propose to balance the samples and train the machine learning models across two validation strategies: k-fold crossvalidation and leave-oneout validation. Various models have been generated to extract different information from the dataset. The performance of the models across the different strategies has been reported. The current research study offers a viable way to maintain consumer trust and advance transparency in the honey sector, in addition to helping to protect the purity of honey products. Hence, by prioritizing food safety and quality, smart cities can create healthier, safer, and more resilient communities for their residents
AI vs. Human Programmers: Complexity and Performance in Code Generation
Full text linkLarge language models, like ChatGPT, have shown the ability to do a variety of tasks in different fields, and this has increased efficiency greatly. However, their increasing use is causing concern about the potential job displacement, particularly in the technical fields. While there have been many studies on the performance of large language models in technical fields, there is a notable absence in assessing their performances in programming. This study fills this gap by comparing ChatGPT (GPT-4) and human experts in the coding discipline to determine if ChatGPT has advanced to a point where it can replace human programmers. To accomplish this goal, this study has produced 300 Python programs with ChatGPT (GPT-4) and compared them with functionally equivalent programs written by three experienced human programmers. The evaluation included both quantitative and qualitative evaluations using measures such as Halstead Complexity, Cyclomatic Complexity, and expert judgment by two human evaluators. The results showed statistically significant differences between the ChatGPT-generated and human-written code. Programs that were generated by ChatGPT were shown to be verbose, complex, and resource demanding, which is reflected in higher program volume, difficulty, and cyclomatic complexity scores. In qualitative terms, ChatGPT\u27s code was easier to read, but lagged behind in some key areas, such as the quality of documentation, structuring of functions, and compliance with coding standards. On the other hand, human-written programs performed well in terms of maintainability, error handling, and dealing with edge cases. Although ChatGPT was found to be incredibly efficient at creating working code, the output needed a lot of review and refinement to be considered standard. The study concluded while ChatGPT is a useful tool for generating code, it has not yet reached the level needed to replace human expertise in programming
Sentiment Analysis of Balochi Text Using Deep Learning
Full text linkBalochi is a low-resource language with limited available data for computational modelling. This study aims to perform sentiment analysis on Balochi text using machine learning techniques. To address the scarcity of linguistic data, we contribute a large, newly constructed dataset of Balochi text. Our proposed model incorporates feature extraction and data augmentation within deep learning algorithms to classify sentiments as positive, negative, or neutral. We evaluate both traditional machine learning methods—such as Random Forest and Support Vector Machine (SVM)—and advanced deep learning models, including Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU).The experimental evidence proves that LSTM and GRU are more effective than traditional methods, and the accuracy rates of sentiment classification with their help are 83.57% and 81.23%, respectively. It has been experimentally verified that, when it comes to the Balochi sentiment analysis, deep learning methods can be more effective than the traditional ones
Water Quality Categorization and Sustainable Management using IoT and Machine Learning.
Full text linkThe critical issue of water deficiency is worsened by widespread wastage in both domestic and industrial sectors globally. A survey conducted by some international organizations indicates that only Pakistan is wasting ten trillion gallons of water. To address this challenge, our study introduces a comprehensive water usage optimization system by using the machine learning opportunities to help improve the classification and prediction of water quality (WQ). Also, using the sensor meters to collect real-time water quality (WQ) parameters in order to facilitate categorization of water sample. The real-time data collected from the sensor meters would be transferred in the storage for further analysis and usage. Several machine learning algorithms are used for the classification. The classifiers include Random Forest (RF), Logistic Regression (LR), Decision Tree (DT), and K-nearest neighbors (KNN). The dataset used for the categorization of water consists of 3277 samples. The water quality parameters used for the categorization were water pH, Turbidity, and Total dissolved solids (TDS) values. The stacked classifiers take the water parameters and provide an appropriate approach for forecasting water quality as accurately as possible. Beyond classification, our system provides targeted recommendations for recycling and reusing water based on its categorized quality. These recommendations are tailored to support sustainable water practices at both household and industrial levels. By combining real-time sensing with intelligent prediction and tailored guidance, the proposed system not only enhances water management but also contributes to long-term conservation efforts. This study serves as a step forward in the integration of smart technologies for sustainable resource utilization
Development of a Diagnostic Model for Pancreatic Ductal Adenocarcinoma Using Nature-Inspired Optimization Algorithm and Machine Learning Techniques
Full text linkPDAC is one of the most harmful cancer causes due to late diagnosis, its rapid progression, and an 11% survival rate of 5 years. Current methods for diagnosis are very costly, uncomfortable, and unreliable, However, better and more accurate solutions are needed. This study proposes a diagnostic model using urinary biomarkers and machine learning techniques for early detection. Key urinary biomarkers, including LYVE-1, REG1B, TFF1, and plasma CA19-9 are used with patient data. Particle Swarm Optimization is used here for feature selection and hyperparameter tuning, optimizes the machine learning classifiers like Support Vector Machine, Logistic Regression, and Random Forest. Accuracy, precision, recall, and F1-score are used as evaluation metrics; however, random forest achieves the highest accuracy of 89.83%. This study shows how PDAC detection changes after combining molecular diagnostics with machine learning. Future research could explore the study of hybrid swarm intelligence algorithms and increase the data set to make further enhancements to diagnostic capabilities. This model shows a great step toward a quick and accurate diagnosis of PDAC and improves patient outcomes and survival rates
Centralized and Decentralized Approach to Monsoon Precipitation Forecasting in Pakistan
Full text linkRainfall, is one of the most important meteorological factors that affects many parts of our everyday lives including crop productivity, water quality, livestock availability, hydroelectric power generation to name a few. Rainfall prediction can significantly contribute to boosting the economy by enabling better planning, risk management, and resource allocation in various industrial sectors. In this study, forty years of monsoon precipitation data is gathered for 39 stations across five zones in Pakistan. We propose a multi-step Long Short-Term Memory (LSTM)-based prediction model capable of forecasting Monsoon yearly data. Three LSTM models stack, bidirectional and convolutional are applied on the dataset and the performance of these models are analysed using a centralized and a decentralized approach. It is observed that the RMSE score of the LSTM models across the centralized strategy was found better than the decentralized approach, whereby 100% of the models in the centralized had a lower RMSE as compared to the decentralized one. Moreover, in the centralized approach 78.7% of the models across the different zones exhibited R2 > 0.9 values indicating a general fit to the model
Advancing Agriculture with IoT and a Smart Fertilizer Recommendation System
Full text linkAgriculture is a key contributor to Pakistan’s GDP, and optimizing fertilization is crucial for enhancing crop yield and ensuring food security. This research presents a real-time, IoT-based soil analysis model that replaces traditional off-site testing, providing instant and site-specific fertilizer recommendations. The system integrates an IoT-enabled device to assess soil nutrient levels and employs a regression algorithm to predict the required NPK quantities. A realistic soil dataset is used to train and validate the model, ensuring accurate predictions. With an 88-92% accuracy rate, the system effectively recommends fertilizers, enabling precision farming and optimizing resource utilization. This reduces reliance on conventional soil testing methods, minimizing fertilizer wastage and improving soil sustainability. The real-time analysis supports data-driven farming decisions, ensuring balanced nutrient application and promoting sustainable agricultural practices. Additionally, this innovation aligns with the Sustainable Development Goals (SDGs) by modernizing agricultural techniques, enhancing food security, and supporting economic growth in farming communities.The IoT-based smart fertilizer recommendation system offers a cost-effective, accurate, and sustainable solution to improve agricultural productivity and promote precision farming.