VFAST - Virtual Foundation for Advancement of Science and Technology (Pakistan)
Not a member yet
1255 research outputs found
Sort by
SafeCon: AI-Powered Real-Time Cyber Grooming Detection System
Full text linkIn recent times, the rise in online communication has unfortunately led to a significant increase in harmful activities. Countless instances involve people, especially children, becoming victims of distressing experiences like online sexual conversation. Reports suggest that a substantial number of young individuals, approximately one in four, have encountered online harassment or inappropriate content. Additionally, there has been a disturbing surge in cases involving the exploitation of children through grooming and exposure to explicit content. Leveraging the PAN12 dataset, we employ the Universal Sentence Encoder (USE) to generate text embeddings, reduce dimensionality with Principal Component Analysis (PCA), and apply K-means clustering with an optimal number of clusters determined by the Silhouette Score. This approach identifies sexually predatory conversation, enabling real-time moderation to protect users. The system also evaluates performance using a manually labeled dataset, ensuring robust detection of harmful content
Sentiment-Aware Summary Generation for User Reviews Using Deep Learning Models
Full text linkUser reviews are a valuable source of information for potential buyers. User reviews provide valuable feedback that reflects genuine customer experiences, enabling future buyers to make more informed decisions. Reviewing every single comment can be a challenging, lengthy, and overwhelming process. Text summarization helps by providing concise summaries, allowing users to quickly grasp key points from multiple reviews. Additionally, sentiment analysis extracts subjective information, such as overall opinions, strengths, weaknesses, and recommendations, helping potential buyers in making informed decisions. This study proposes the XLNet model for sentiment analysis of user reviews and LED (Longformer Encoder-Decoder) for text summarization of user reviews. XLNet is a pre-trained transformer model that captures bidirectional relationships between words to improve sentiment analysis, while LED is a transformer-based model designed for efficient text summarization of long documents by using a sparse attention mechanism. The results show that XLNet achieves high accuracy in sentiment classification on the Amazon Fine Food Reviews dataset of 50K reviews, while LED generates effective and concise summaries, achieving the highest ROUGE and METEOR scores among the tested summarization models. To ensure robust evaluation, multiple metrics, including ROUGE, METEOR, MoverScore, and BERTScore-F1, were applied across models, providing both lexical and semantic perspectives on summarization quality. By performing both sentiment analysis and text summarization within a single framework, this approach efficiently extracts meaningful insights from large datasets, streamlining the decision-making process for users
Enhancing Multivariate Data Classification Using Graph Convolutional Networks: A Comparative Evaluation with PCA and t-SNE
Full text linkGraph Neural Networks (GNNs) have become a critical tool for learning on structured data, particularly for their ability to capture feature dependencies within multivariate datasets. This study investigates the application of Graph Convolutional Networks (GCNs) to improve classification performance on such data. We constructed a fully connected graph using cosine similarity to compare our GCN approach against standard dimensionality reduction techniques like PCA and t-SNE. In our tests, the GCN model clearly surpassed these baselines in accuracy, precision, recall, and F1-score.But the method has its own trade-offs: dense graphs are computationally costly, there are chances of overfitting on smaller datasets and the resulting embeddings may be uninterpretable. We cannot make such a claim of generality at this point because this review has only limited itself to one multivariate data set. In future endeavour we would wish to compare GCNs to more powerful classifiers, such as Random Forests, Deep Neural Networks, with greater regard to efficiency in training and resistance to noise. To summarize, this paper confirms that GNNs are an ideal option when it comes to multivariate classification, although scalability and interpretability have to be resolved in order to make it applicable in the real-life context
Interpretable Physics-Informed Neural Network for Reliable Humidity Forecasting in Contrasting Climates of Sindh.
Full text linkReliable forecasting of humidity is fundamental to effective climate monitoring, agricultural scheduling, and public health planning. Traditional machine learning frameworks, however, struggle to mirror the largely nonlinear climatological processes governing moisture distribution and rarely satisfy governing physical laws. This work introduces a physics-informed neural network (PINN) explicitly tailored to forecast daily mean humidity, embedding thermodynamic principles and smaller-scale climatological knowledge to bolster generalization and interpretability. The analysis is based on a comprehensive dataset of meteorological measurements from three climatically contrasting districts in Sindh, encompassing temperature, wind velocity, cumulative rainfall, and derived interaction terms. The PINN topology augments the mean squared error loss with a physically grounded rainfall–humidity relationship, thereby constraining the output to physically realizable humidity profiles. To decode the model’s inner operation, we employed SHAP and LIME, providing affected stakeholders with a transparent rationale for predictions. Comparative benchmarks reveal that the physics-informed architecture exceeds the forecast skill of conventional regression, ensemble methods, and standard feed-forward neural networks, securing R² metrics greater than 0.99 and uniformly low mean squared error across all three locales. Sensitivity assessments based on SHAP and LIME indicate that seasonal modulation and the interaction between wind and humidity are the predominant drivers of the forecasts, whereas the contribution of rainfall-derived predictors remains negligible. The combination of PINN with contemporary interpretability methodologies yields a forecasting paradigm for meteorology that is both transparent and governed by established physical laws.
Investigation and Analysis of the Role of Leaf’s Vein Features for Automated Classification of Plants
Full text linkSeveral researchers have proposed various methods to classify plants. Veins pattern of leaf contains imperative information for plants identification regardless its complex modality. This research describes a methodology for classifying different types of plants using various features i.e. leaf shape, texture, color and newly proposed vein features. The proposed algorithm comprises of three distinct stages that are preprocessing, feature extraction and classification. The characteristics of plant can be determined from leaf structures. The classification approach used in this research based on plant leaves as leave belongs to different plants has different characteristics. The used dataset for the proposed approach is a flavia database that contains 32 kinds of plant leaves. It is observed from experimental results that the classification accuracy of 97.25% was achieved using a Random Forest classifier when a combination of shape, HSV color moment, texture and venation features are used
IoT-Enabled Machine Learning Framework for Precision Agriculture: Achieving Near-Perfect Crop Yield Prediction in Pakistan\u27s Diverse Agro-Climatic Zones
Full text linkThis study introduces an integrated IoT-based framework for machine learning of agricultural crop yield prediction without rivaling in the accuracy levels. After an intensive survey of 16 machine learning algorithms on multiyear data from the principle agricultural zones of Pakistan, we show that ensemble models are capable of achieving close to perfect prediction accuracy. The Random Forest method had excellent performance with R² = 0.99, RMSE = 538.37 kg/ha and MAPE = 0.2944% which are a new benchmark in the field of agricultural yield prediction. Analysis of 400 agricultural records across seven districts showed evident outperformance of the tree-based ensemble methods against conventional ones. The proposed end-to-end framework integrates IoT sensor networks and cutting-edge ML models to monitor in real-time, and with high accuracy, the yields of major crops viz. wheat, cotton, sugarcane and rice. Here, we present an important leapfrogging in precision agriculture technology development that provides disruptive potential to the agricultural planning, resource allocation and food security improvement of developing countries by fusing IoT and machine learning
Emotion Detection and Analysis using Textual Data through Trainable and Pre-trained Word Embedding Methods
Full text linkEmotion expression modes play a significant role in human communication. Humans use emotions to convey their state of mind to each other on platforms such as X (formerly Twitter), Facebook, and other online social networks. People often express their emotions using free text, which triggers a vast research area of emotion detection and analysis. This work aims to detect and analyze emotions from unstructured text data. For this purpose, this research study proposes a solution to the problem by building a deep artificial neural network model using trainable and pre-trained word embedding methods. Afterward, the performance of the models developed with different word embeddings is evaluated using the performance metrics. Experimental works demonstrate that the deep artificial neural network with trainable word embedding surpassed all other models by achieving 67.36% accuracy, 53.27% recall, 82.62% precision, and 64.50% F-measure
A Novel Blockchain Proof of Validation Scheme Based on Explainable AI for Healthcare Workload
Full text linkThese days, the usage of blockchain with machine learning to optimise data validation in terms of transparency, validity, and immutability has been increasing daily. Therefore, many complex applications, such as healthcare and related disease processes, have recently required the implementation of many remote resources in a transparent form. The blockchain provides real-time security validation based on proof of work validation schemes. To understand the dynamic situation of blockchain, mainly machine learning implemented for the decision and improve the efficiency of the security. However, there are many limitations when using blockchain technology with machine learning. Therefore, to cope with this issue, a novel blockchain proof of validation scheme based on explainable AI for healthcare applications is needed to process the decision of blockchain with machine learning in a more explainable way. We present the blockchain proof of work validation explainable AI (PoWV-XAI) to control the delay, energy, cost and security dynamic issues compared to existing blockchains with machine learning algorithms. The proposed PoWV-XAI algorithm suggested different metaheuristic schemes and supported the explainability of healthcare workload execution on other nodes, such as local and server. Simulation results show that the proposed PoWV-XAI is more explainable, and all decisions, such as processing delay, validation, security, energy, and cost, are explainable compared to existing blockchain methods
A novel Primary Key infrastructure IoT enabled secure Access Control Framework Based smart home applications
Full text linkThis paper presents a comprehensive security framework for smart home environments, integrating advanced authentication, access control mechanisms, and machine learning for robust IoT protection. A typical smart home ecosystem is illustrated with interconnected devices—air conditioning systems, robotic assistants, surveillance cameras, and thermostats—linked through a centralised wireless network that facilitates seamless communication and remote control via the internet. The framework emphasises secure authentication and authorisation processes, using public key infrastructure (PKI) to validate devices and users while issuing, renewing, and revoking certificates for encrypted communication. The mathematical model outlines device and user authentication, validation functions, and access control mechanisms to ensure secure operations. Fine-grained access control is implemented to grant permissions based on specific conditions, ensuring flexible yet secure resource allocation. Communication security is maintained through encryption and decryption, safeguarding data transmitted across devices and networks. To mitigate security risks, a convolutional neural network is employed for anomaly detection, identifying threats by recognising deviations in regular patterns. Additionally, the framework addresses interoperability by adhering to standard-compliance protocols, facilitating seamless integration across diverse devices. Resource optimisation techniques are introduced to maximise efficiency based on the number and capabilities of devices in the network. User interaction is streamlined through an intuitive interface that supports secure and user-friendly system access. The proposed SAC-PKI algorithm serves as the foundation for the framework, detailing sequential steps for authentication, certificate management, access control, and anomaly detection. By leveraging adaptive security features and advanced threat detection, this framework provides a robust solution for enhancing the cybersecurity of smart home deployments, addressing vulnerabilities, and ensuring efficient resource utilisation in IoT environments
Sentiment Analysis of Multilingual Roman Text for E-Commerce Reviews using Machine Learning Approaches
Full text linkSentiment analysis, a type of natural language processing (NLP) analyzes the text data to extract and identify subjective information including attitudes, opinions, and feelings. Sentiment analysis can be used to examine audience feedback and reviews in the context of multilingual product reviews. In this paper, a sentiment analysis model using machine learning approaches has been developed for multilingual product reviews in Roman Urdu or Sindhi to determine how the public feels about certain posts, products, etc. The importance of sentiment analysis for product context reviews in many languages in Roman is multifaceted. It can offer insightful information on the preferences of the likes and dislikes of the audience. To accomplish multilingual sentiment analysis, a dataset of reviews in Roman Urdu and Sindhi languages from diverse online platforms and social media sources like YouTube, Facebook, TikTok, Daraz, and Instagram was collected.To identify pertinent features essential for categorizing reviews into negative, positive, or neutral sentiments based on polarity, the Term Frequency Inverse Document Frequency (TF-IDF) method was used. For classification, five different machine learning classifiers including Linear Regression (LR), Naive Bayes (NB), Support Vector Machine (SVM), Random Forest (RF), and K-nearest neighbors (KNN) were used. The classification results were measured in terms of precision score, recall score, and F1-score. With TF-IDF, the SVM, and LR outperformed than other classifiers and obtained an F1-score of 0.77%, and 0.78%. To further improve the classification accuracy, the Synthetic Minority Over-sampling TEchnique (SMOTE) was used to manage the class imbalance problem. With SMOTE, the classification accuracy of LR and SVM was improved to 0.79% and 0.80%