PhD

Permanent URI for this community

https://escholar.umt.edu.pk/handle/123456789/4415

Browse

Browsing PhD by Issue Date

Filter results by year or month
Now showing 1 - 20 of 20
  • No Thumbnail Available
    Item
  • No Thumbnail Available
    Item
    A Computationally Intelligent System for Prediction of Protein Function using Pattern Recognition
    (UMT,Lahore, 2020) AHMAD HASSAN BUTT
    Pattern recognition systems are emerging in a multitude of computer applications. They have been part of many computationally intelligent systems like optical character recognition systems, biometric verification systems, weather forecasting, decision support systems, etc. Computationally intelligent systems are also considered significant components in the toolkit of a biologist. Such systems are essentially required by the majority of the modern research projects in the biological sciences. Most of these projects use computationally intelligent systems for either DNA or protein sequence analysis. Protein molecules are composed of a large sequence of Amino Acids. With the rapid discovery of new protein sequences in past decades, functional identification of the hypothetical or uncharacterized protein sequence or its primary structure is confronted as a challenging task in computational biology and proteomics. To explore the problems associated with protein function prediction, some computational techniques were proposed in the past, but are still not effective in terms of efficiency and accuracy. Based on pattern recognition feature extractions and machine learning classification algorithms, the outcome of the current research study has developed a computationally intelligent system that will be an effective practical approach in predicting protein functional attributes. The observed results, obtained from the proposed system in predicting protein functions, have shown better performance outcomes in terms of accuracy as compared to the existing state-of-art systems. Finally, we conclude that the proposed system will be effective and useful in problems relating to Bioinformatics, medicinal biology and drug discovery. This system will enhance the experimental research dynamics into exceptionally interpreting and analyzing the biological data for quick progress in areas of vaccine discovery, drug interactions, disease predictions and most importantly biological datasets characterizations.
  • No Thumbnail Available
    Item
    BRIDGING THE COMMUNICATION BARRIER FOR THE DEAF IN PAKISTAN USING INFORMATION TECHNOLOGY
    (UMT, Lahore, 2020-04) NABEEL SABIR KHAN
    The deaf community in the world uses a gesture-based language, generally known as sign language. Every country has a different sign language; for instance, USA has American Sign Language (ASL) and UK has British Sign Language (BSL). The deaf community in Pakistan uses Pakistan Sign Language (PSL), which like other natural languages, has a vocabulary, sentence structure, and word order. Majority of the hearing community is not aware of PSL due to which there exists a huge communication gap between the two groups. Similarly, deaf persons are unable to read text written in English and Urdu. Hence, the provision of an effective translation model can support the cognitive capability of the deaf community to interpret natural language materials available on the Internet and in other useful resources. This research involves exploiting natural language processing (NLP) techniques to support the deaf community by proposing a novel machine translation model that translates English sentences into equivalent Pakistan Sign Language (PSL). Though a large number of machine translation systems have been successfully implemented for natural to natural language translations, natural to sign language machine translation is a relatively new area of research. State-of-the-art works in natural to sign language translation are mostly domain specific and suffer from low accuracy scores. Major reasons are specialized language structures for sign languages, and lack of annotated corpora to facilitate development of more generalizable machine translation systems. To this end, a grammar-based machine translation model is proposed to translate sentences written in English language into equivalent PSL sentences. To the best of our knowledge, this is a first effort to translate any natural language to PSL using core NLP techniques. The proposed approach involves a structured process to investigate the linguistic structure of PSL and formulate the grammatical structure of PSL sentences. These rules are then formalized into a context-free grammar, which, in turn, can be efficiently implemented as a parsing module for translation and validation of target PSL sentences. The whole concept is implemented as a software system, comprising the NLP pipeline and an external service to render the avatar-based video of translated words, in order to compensate the cognitive hearing deficit of deaf people. The accuracy of the proposed translation model has been evaluated manually and automatically. Quantitative results reveal a very promising Bilingual Evaluation Understudy (BLEU) score of 0.78. Subjective evaluations demonstrate that the system can compensate for the cognitive hearing deficit of end users through the system output expressed as a readily interpretable avatar. Comparative analysis shows that our proposed system works well for simple sentences but struggles to translate compound and compound complex sentences correctly, which warrants future ongoing research.
  • Thumbnail Image
    Item
    A COGNITIVE FRAMEWORK FOR THE IMPROVEMENT OF LEARNING ANALYTICS EFFICIENCY IN INTRODUCTORY PROGRAMMING COURSES
    (UMT, Lahore, 2021) Uzma Omer
    Learning analytics (LA) has become a popular discipline among educationists and researchers as it has a potential to reveal new facets of teaching and learning that could be utilized to improve the efficiencies of related learning environments. Introductory programming courses (IPCs) hold special significance as these courses lay down the foundation for subsequent higher level courses in computer science and associated disciplines. The LA studies in IPCs are mostly anecdotal as less or no attention is given to examine learning at various cognitive levels. This research is designed to find improvements in learning analytics in IPCs by evaluating the cognitive aspects of students’ learning. It aims to explore more granular technique of LA that could lead to enhance the efficiency of LA in IPCs. The objectives of this work are addressed by proposing a framework for cognitive learning analytics in IPCs which serves as a platform that provides structure to the concept data using the technique of concept mapping and examines proliferation of cognitive learning on related concepts using assessment data. The framework is evaluated by predicting performance of learners on a number of IPC concepts through the metrics established from cognitive maps of learners, acquired by deploying the related layers of framework. It was identified that performance predictions through proposed metrics helped in improving efficiency of learning analytics performed in existing work. The research is concluded by presenting prediction accuracies acquired while evaluating the framework which are comparable to the related studies where the proposed technique showed better accuracies as compared to most of the related work. This work contributes by proposing a framework of cognitive learning analytics in introductory programming courses and presenting metrics to measure the cognitive performances which predict the learning performances with improved accuracies.
  • Thumbnail Image
    Item
    A framework for the prediction of earthquakes in western himalayas
    (UMT, Lahore, 2021) RABIA TEHSEEN
    Earthquake is a natural phenomenon which may cause significant loss of life and infrastructure. Researchers have applied multiple artificial intelligence-based techniques to predict earthquakes, but they have not been able to achieve high accuracies due to the huge size of multidimensional data, communication delays, transmission latency, processing capacity limitations and data privacy issues. In this research, a novel earthquake prediction framework has been proposed that is capable enough to process data using different Artificial Intelligence (AI) approaches. The proposed framework has been equally effective in both centralized and distributed settings. We have initially implemented the proposed framework using Fuzzy Expert System (FES). FES is a traditional benchmark system used in the literature for earthquake prediction. We have achieved 47% earthquake prediction accuracy by using FES. To improve the accuracy in earthquake prediction, the proposed framework has been implemented using different Machine learning (ML) methods and state-of-the-art Federated learning (FL) mechanism. FL has given better performance over already developed ML methods applied to earthquake prediction in terms of efficiency, reliability, and precision. We have tested the proposed framework by analyzing three-dimensional data within 100 km radial area from 34.708o N, 72.5478o E in Western Himalayas. The results of the proposed framework have been validated against instrumentally recorded regional seismic data of last thirty six years, and 88.87% prediction accuracy has been recorded while using FL technique. Implementation of proposed framework can serve as a significant component for developing early warning earthquakes systems.
  • Thumbnail Image
    Item
    Cardiac Disorder Classification by Electrocardiogram Sensing Using Deep Neural Network
    (UMT,Lahore, 2022) ALI HAIDER KHAN
    Electrocardiogram (ECG) is the most common and low-cost diagnostic tool used in healthcare institutes for screening heart electrical signals. The abnormal heart signals are commonly known as arrhythmia. Cardiac arrhythmia can be dangerous, or in most cases, it can cause death. The arrhythmia can be of different types, and it can be detected by an ECG test. The automated screening of arrhythmia classification using ECG beats is developed for ages. The automated systems that can be adapted as a tool for screening arrhythmia classification play a vital role not only for the patients but can also assist the doctors. The deep learning-based automated arrhythmia classification techniques are developed with high accuracy results but are still not adopted by healthcare professionals as the generalized approach. The primary concerns that affect the success of the developed arrhythmia detection systems are (i) manual features selection, (ii) techniques used for features extraction, and (iii) algorithm used for classification and the most important is the use of imbalanced data for classification
  • Thumbnail Image
    Item
    Modeling and Parametric Analysis of Solar Energy Systems
    (UMT,Lahore, 2022-04)
    Renewable energy is becoming important due to increase in global warming caused by fossil fuel based energy generation. World population has signifi- cantly increased in the last two decades and growing industrial revolution has also been happening since 21st century. Hence, energy demand is multiplying due to increasing world population and growing industrial revolution with each passing year. Energy demand is being largely met by conventional fossil fuel energy generation sources, which results in higher greenhouse gas emissions and global warming. Researchers, scientists, and academicians pro- posed renewable energy generations viable for resolving the global warming issue. In renewable energy, solar energy is the most abundant energy source available to cater the issues of meeting global energy demand and reducing greenhouse gas emissions in the environment. Solar energy generation is achieved through photovoltaic panels and solar energy collectors with the help of small particles to capture energy from solar radiations.
  • Thumbnail Image
    Item
    Design and Development of a Searching Technique for Web of Things
    (UMT,Lahore, 2023) MUHAMMAD REHAN FAHEEM
    ith the advancement of Technology, there is an increase in the number of interconnected real- world devices, sensors, actuators and physical devices. All these devices are interconnected through the internet which is known as Internet of Things (IoT). In the near future, research shows that billions or even trillions of devices will be connected to the Internet. By this unexpected increase in the number of devices, researchers are facing difficulties in discovery, describing the things and searching the devices for using them due to the lack of graphical user interface. However few researchers, works on the graphical user interface but still it is a big issue. Nowadays, the researchers have been working on the search engine and web interface to search the things through web interface which is known as the Web of Things (WoT). The ultimate goal of Web of Things is to build an ideal search engine where the user or even devices can find other devices anywhere and at any time for using the resources of other devices.
  • Thumbnail Image
    Item
    Framework for prediction of oncogenomic progression aiding personalized treatment of carcinoma
    (UMT, Lahore, 2023) Asghar Ali Shah
    DNA makes up genes, and each gene has a unique sequence. A mutation is a permanent alteration to the nucleotide collection in DNA that results from recombination or replication within the genetic base and some of the mutations cause cancer. Cancer is a disease in which cells in a particular body component proliferate and replicate in an uncontrolled manner. Most of the earlier research uses images to detect cancer after symptoms start to manifest, which is a late discovery. Therefore, Numerous lives can be saved if cancer is discovered in its preliminary stages. This study proposed an Ensemble Learning (EL) model based on three deep learning models such as Long Short-Term Memory (LSTM), Gated Recurrent Units (GRU), and Bi-directional LSTM (BLSTM) to detect mutation in gene sequences to detect cancer progression in an early stage. The proposed model is implemented on breast, thyroid, lower-grade glioma, sarcoma, and gastric cancer. The driver genes in breast, thyroid, lower-grade glioma, sarcoma, and gastric cancer are 99, 40, 38, 8, and 61, respectively. Different feature extraction algorithms are applied to these gene sequences. The learning approaches are validated and tested using three different testing techniques: self-consistency test, independent set test and a 10-Fold cross-validation test. Then, multiple statistical tools are used to evaluate the performance of the proposed model such as accuracy, sensitivity, specificity, Mathew's correlation coefficient, ROC curve and decision boundary. The proposed study shows the highest accuracy of 99% for the identification of breast adenocarcinoma, thyroid adenocarcinoma, lower glioma, sarcoma and gastric cancer using Bi-LSTM shown in Table XII, XIII, XIV, XV and XVI. None of the previous studies uses ensemble leaning approach for the identification of any type of cancer. The current proposed study is first that focus on this and provide the state of the art results. The ensemble learning approach shows the highest accuracy of 96% for breast cancer, 93% for thyroid cancer, 97% for lower glioma cancer, 80% for sarcoma and 96% for gastric cancer. This is the highest accuracy of the identification of these type of cancer till date. The comparison of the results with the previous studies are explained in Table XVII.
  • Thumbnail Image
    Item
    An intelligent technical and vocational education and training (TVET) course recommendation system based on the trainee’s aptitude
    (UMT, Lahore, 2024) Rana Hammad Hassan
    Personality encompasses the distinct patterns of thoughts, emotions, and behaviors that differentiate individuals and typically remain stable throughout one's life. Aligning these individual traits with learning aptitudes holds promise for improving course outcomes, maximizing returns on investment, and reducing dropout rates significantly. This interdisciplinary research bridges insights from Computer Science (CS) and Human Psychology by analyzing data from Technical and Vocational Education and Training (TVET) programs, focusing on the Big Five Personality traits (BFI). This study marks a pioneering effort in both the Pakistani and global TVET sectors, linking TVET learning skills with individual personalities. We have addressed important ethical considerations, including data privacy and informed consent, ensuring the responsible use of human subjects in this study. The study introduces the deep learning-based Personality-aware TVET Course Recommendation System (TVET-CRS). Over four years, data collection, analysis, and evaluation were conducted in collaboration with one of Punjab, Pakistan's largest TVET training providers. To test the hypothesis, machine learning techniques such as Chi-Square analysis, 5-fold cross-validation, and ensemble classifiers were employed. These methodologies laid the foundation for the development of a robust personality-aware TVET course recommendation system. Notably, the research's original contributions include the development of TVET-CRS and the creation of the first personality dataset specific to Pakistan's TVET sector. TVET-CRS has achieved an impressive accuracy rate of 91%, surpassing benchmarks established in existing literature across all evaluation metrics. Particularly noteworthy is its Cohen’s Kappa score of 0.84, indicating substantial agreement between predicted and actual trades, alongside its lowest error rate of NMAE at 0.04 and highest ranking NDCG of 0.96. These findings carry significant implications across various stages of the TVET training cycle, including dropout prediction, career guidance, on-the-job training assessments, exam evaluations, and course recommendations. The adaptable methodology employed in this research holds potential for application within the TVET sectors of developing countries. The dissemination of these findings is anticipated to attract interest from TVET training providers, policymakers, international funding agencies, researchers, and academics alike, aiming to enrich knowledge, enhance the effectiveness of TVET programs, and maximize returns on investment.
  • Thumbnail Image
    Item
    Integrated real-time distributed stream-disk processing architecture for un-structured big data
    (UMT, Lahore, 2024) Erum Mehmood
    Real-time ETL(Extract-Transform-Load) is a crucial component of the grow ing demand for quicker business decisions aimed at numerous contemporary applications. The foundation of real-time ETL is un-structured data stream extraction from multi-source and transformation employing distributed disk data because of the volume and velocity of the data. Heterogeneity is another key aspect of complex networks and smart devices for using it as nature of live streams. The heterogeneous stream-disk join is a significant research topic in real-time processing applications because it can directly affect the data ana lytics. Multiple issues, including stream loss, scalability, disk access cost, and data accuracy, should be considered during heterogeneous stream-disk join transformation. As a result, developing an architecture for fundamental ETL building blocks in real time continues to be quite challenging. In this work, two architectures are proposed which can help organizations in improving their current decision making systems. It is particularly focused on speeding up stream-disk joins (transformation), which are the most expensive operation in stream processing because these require frequent disk access. This thesis presents its first architecture, without having to worry about the format of the data sources, for real-time ETL that would convert the unstructured stream of data after combining it with distributed disk data. First proposed architecture is capable to perform analytics on heterogeneous data without loosing the flexibility of data from multiple sources at native speed. To overcome the issue of heterogeneous un-structured stream disk join, second architecture of this thesis is proposed: an integrated distributed het erogeneous stream-disk join architecture DHSDJArch which can prevent stream data loss as well as maintaining balance between heterogeneous dis tributed data sources and accuracy of stream-disk join. This four phased dis tributed architecture is proposed for the multi-objective optimization to trans form heterogeneous incomplete stream. To prevent stream loss, configuration of log retention is proposed based on the characteristics of distributed event streaming platform (DESP). Specifically, two transformations are proposed to xi pre-process heterogeneous streams and to join pre-processed stream with dis tributed disk data by performing real-time disk access while compensating the differences between data sources and streaming application, respectively. Additionally, a cutting-edge data pipeline is described for stream-disk join that makes use of partition-based input and a best-effort in-memory database strategy to lessen the number of times the disk is accessed. The suggested architectures deal with problems including real-time processing in dispersed environments, heterogeneous streams, ignored un-matching streams, disk over head, and stream data loss in streams. On both local and distributed worksta tions, experimental results utilising a stream generator and real-world datasets demonstrate that the proposed architectures greatly enhance throughput, es pecially for high numbers of stream tuples with huge datasets. According to experimental results, throughput scaling is linear with respect to the quan tity of input streams and dataset sizes. Performance criteria considered in this study corroborate the functioning of proposed architectures in terms of accuracy, log retention policy, scaling, stability and cloud data storage. Two contributions are being made in this work by developing and evaluat ing two architectures focusing on unstructured big data generated from homo geneous or heterogeneous multi-sources. Rigorous evaluation shows that there are no existing architectures that dominate overall performance of real-time distributed stream processing under the conditions of un-structured heteroge neous big data streams.
  • Thumbnail Image
    Item
  • Thumbnail Image
    Item
    A Robust Ad-hoc Collaborative Space for Customized Mobile Application Development
    (UMT,Lahore, 2024) Imran Abbas Khawaja
    By combining the characteristics of mobility, computing, storage, and communication, portable electronic devices like smartphones, wearables, tablets, etc. have completely changed the way we live. The software applications for these mobile devices possess the capability to disseminate and receive information, thereby enabling the development of applications that may be executed in a collaborative manner. However, in a situation of a critical nature which includes unavailability of infrastructure, communication breakdown, or disaster, information sharing across these devices may not be consistently possible, which causes a variety of problems for the applications running in the collaborative spaces. Thus, there is a need to facilitate the developers of collaborative applications with a framework to build more robust applications without worrying about intermittent communication failures. As a contribution, this research proposes a robust framework that facilitates the development of robust, reliable, and efficient collaborative applications that communicate across devices while working in an ad-hoc environment. Whereby, the proposed framework provides an abstraction of some major components including service exposition, service registration, storage, and synchronization in a robust manner. Thus, enables the application developers to design and build customized applications as per their needs without worrying about intermittent failures, which are taken care of by different components of the framework. The efficacy of the framework has been demonstrated through a detailed experimental evaluation using a custom-developed collaborative application run in a variety of operational settings over different portable devices in ad-hoc settings.
  • Thumbnail Image
    Item
    A multiclassification deep framework for the diagnosis of skin cancer using dermoscopy images
    (UMT, Lahore, 2024) Ahmad Naeem
    Skin cancer is a common cancer affecting millions of people annually. Skin cells inside the body that grow in unusual patterns are a sign of this invasive disease. The cells then spread to other organs and tissues through the lymph nodes and destroy them. Lifestyle changes and increased solar exposure contribute to the rise in the incidence of skin cancer. Early identification and staging are essential due to the high mortality rate associated with skin cancer. Furthermore, differentiating the specific categories of skin cancers, such as melanoma (Mel), melanocytic nevus (Mn), basal cell carcinoma (bcc), squamous cell carcinoma (Scc), benign keratosis (Bk), Actinic keratosis (Ak), Dermatofibroma (Df) and Vascular lesion (Vl) is also necessary. Thus, two novel methods were designed for the classification of several categories of skin cancer. Firstly, a deep learning-based method named DVFNet for the detection of skin cancer from dermoscopy images is designed. To detect skin cancer images are pre-processed using anisotropic diffusion methods to remove artifacts and noise which enhances the quality of images. A combination of the VGG19 architecture and the Histogram of Oriented Gradients (HOG) is used in this research for discriminative feature extraction. SMOTE Tomek is used to resolve the problem of imbalanced images in the multiple classes of the publicly available ISIC 2019 dataset. This study utilizes segmentation to pinpoint areas of significantly damaged skin cells. A feature vector map is created by combining the features of HOG and VGG19. Multiclassification is accomplished by CNN using feature vector maps. A comparison is performed between the DVFNet, benchmark classifiers and state-of-the-art classifiers. Secondly, a newly developed deep learning model that makes use of two advanced artificial intelligence techniques, Xception ResNet101 (X_R101) for the identification of Mel, Mn, bcc, Scc, Bk, Ak, Df, and VI. To evaluate the performance of the proposed model (X_R101), three publicly available datasets (PH2, DermIS, and HAM10000) are utilized. A comparison is performed between the X_R101 and four benchmark classifiers: MobileNetV2 (BM1), DenseNet201 (BM2), InceptionV3 (BM3), and ResNet50 (BM4) and state-of-the art classifiers. The implementation of borderline SMOTE with X_R101 improves performance substantially. The DVFNet model achieves an accuracy of 98.32% on the ISIC 2019 dataset. Analysis of variance (ANOVA) statistical test is used to validate the DVFNet model accuracy. The X_R101 model attains a prediction accuracy of 98.21%. McNemar statistical test is used to validate the X_R101 model accuracy. The accuracy and effectiveness of the proposed models such as DVFNet and X_R101 provide benefits to dermatologists and other healthcare practitioners in terms of timely identification of skin cancer.
  • Thumbnail Image
    Item
    Multi-Modal Deep Learning Methods for Classification of Chest Diseases Using Different Medical Imaging and Cough Sounds
    (UMT,Lahore, 2024) Hassaan Malik
    st disease encompasses conditions that affect the lungs such as COVID-19, lung cancer (LC), consolidation lung (COL), and many others. Medical professionals are often misled when diagnosing chest disorders. This is due to overlapping symptoms such as fever, cough, sore throat, and other common symptoms. Furthermore, chest X-rays (CXR), cough sounds, and computed tomography (CT) scans are utilized by researchers and medical professionals when diagnosing chest disorders.
  • Thumbnail Image
    Item
    Language resource and model for intrinsic plagiarism detection for urdu language
    (UMT, Lahore, 2025) Muhammad Faraz Manzoor
    In the evolving field of natural language processing (NLP), plagiarism detection has become an essential task, particularly for low-resource languages like Urdu. This PhD research addresses the critical challenge of intrinsic plagiarism detection in Urdu texts by employing a novel framework that combines machine learning, deep learning, and language models. The study conducts a comprehensive analysis at both the paragraph and sentence levels to advance the detection of intrinsic plagiarism. At the paragraph level, a set of 43 stylometry features across six granularity levels was meticulously curated to capture linguistic patterns indicative of plagiarism. The selected models include traditional machine learning techniques such as Logistic Regression, Decision Trees, Support Vector Machines (SVM), K-Nearest Neighbors (KNN), Naive Bayes, Gradient Boosting, and Voting Classifier, alongside deep learning models like GRU, BiLSTM, CNN, LSTM, and MLP, as well as Large Language Models (LLMs) such as BERT and GPT-2. Two distinct experiments were conducted: the first utilized the entire dataset for classification into intrinsic plagiarized and non-plagiarized documents, while the second categorized the dataset into three topical types—Moral Lessons, National Celebrities, and National Events. The Random Forest Classifier achieved an exceptional accuracy of 98.81% in the first experiment, while the Extreme Gradient Boosting Classifier reached an overall accuracy of 99.00% in the second experiment, demonstrating superior capability in distinguishing nuanced stylistic features across different topics. At the sentence level, the study focuses on leveraging various embeddings, including TF IDF, Word2Vec, FastText, and GloVe, in conjunction with machine learning and ensemble learning classifiers. A dataset comprising 2520 balanced documents was used to evaluate the efficacy of these models. The experiments showed promising results, with FastText embeddings combined with Support Vector Classifier and Random Forest emerging as top performers, achieving accuracy viii scores of 0.89. While BiLSTM also demonstrated competitive performance with an accuracy of 0.75, the BERT model underperformed with an accuracy of 0.65, highlighting the challenges of applying LLMs in low-resource languages like Urdu. This research highlights the effectiveness of tailored stylometry features and traditional machine learning models over deep learning and LLMs for intrinsic plagiarism detection in Urdu. The findings underscore the potential for further advancements through the expansion of datasets and the development of more sophisticated language models tailored to the linguistic characteristics of Urdu.
  • Thumbnail Image
    Item
    Reduction of imbalanced data to improve the accuracy of deep learning algorithms for federated learning techniques
    (UMT, Lahore, 2025) Momina Shaheen
    Federated learning is a leading machine learning paradigm that facilitates collaborative model training across decentralized nodes while ensuring data privacy and security. In edge computing environments addressing imbalanced training data is a critical challenge due to its non-independent and identically distributed form and variable size. This research explores the impact of global data imbalance on Federated Learning (FL) model accuracy, revealing complexities in mitigating its negative effects. Through empirical analysis and theoretical investigations, new insights into the mechanisms degrading FL accuracy are uncovered, leading to the proposal of a novel method tailored for FL networks. The proposed framework employs two strategies: global distribution data augmentation and synthesis for rebalancing training data, and client rescheduling by mediators for partial equilibrium among edge devices. Experiments on various distributed datasets reveal significant improvements in learning accuracy. This study's main contribution is its analysis of the negative impact of imbalanced training data on federated learning (FL) model accuracy and the development of effective strategies to mitigate this issue. By integrating AI techniques like data augmentation and class estimation into the FL framework, the approach enhances accuracy with minimal computational overhead. This innovative approach utilizes advanced artificial intelligence methodologies within the federated learning (FL) framework to address imbalanced training data and improve the robustness of FL systems in edge computing. Rigorous experimental validation on two datasets— Fashion-MNIST and a dataset stock data—shows that the method achieves nearly 92% accuracy across both types, highlighting its effectiveness in FL for edge computing viii environments. This experimentation on distinct type of datasets including image classification and financial predictive analytics, the method shows significant enhancements in FL model accuracy, underscoring its potential to revolutionize FL methodologies and foster resilient machine learning (ML) systems in edge computing.
  • Thumbnail Image
    Item
    A framework for the improvement of distributed agile software development based on blockchain
    (UMT, Lahore, 2025) Junaid Nasir Qureshi
    The goal in today’s global software industry is software development using Agile in a distributed environment where teams work across different geographic locations. However, the traditional framework, which is responsible for coordinating, communicating and collaborating in Distributed Agile Software Development (DASD) hasn't encountered the areas such as security, transparency, traceability, and strong teamwork of individuals over distributed geographic locations. Typically, these deficiencies result in immense problems, that include delays in software development and deployment, project failures, unsuccessful contracts with clients and developers, and clients’ dissatisfaction with the way the software was developed in a distributed environment. This research study therefore introduces a novel framework which implements Blockchain technology as a Distributed Agile Software Development (DASD) approach to overcoming these challenges. On a private Ethereum blockchain, smart contracts are used to automate and secure different aspects of the Distributed Agile Software Development processes. Processes such as verifying requirements, organizing tasks as per priority, managing sprint backlogs, developing and creating user stories, testing for acceptance against user stories, transaction security automation and payments that are disseminated to development teams through digital wallets are managed by these smart contracts. Moreover, smart contracts automatically impose penalties on customers who aren't paying on time or for missing payments, and developers for failing to meet their deadlines. However, to address the scalability constraints typically presented in blockchain technology, this research describes the use of the Interplanetary File System (IPFS) as an off-chain storage solution. This integration of IPFS allows for efficient management of large amounts of data without overloading the blockchain. Furthermore, vii the experimental results of the research indicate that this innovative method significantly enhances teamwork synchronization, communication, traceability, transparency, security, and confidence among clients or customers and developers involved in Distributed Agile Software Development. (DASD).
  • Thumbnail Image
    Item
    An intelligent diagnostic model to predict disease associated biomarkers in genomic sequences
    (UMT, Lahore, 2025) Ayesha Karim
    Objective: Cell mutation refers to changes in the genetic material (DNA or RNA) of a cell that can disrupt normal protein synthesis and cell function. While some mutations have minimal effect, others can lead to the production of abnormal or dysregulated proteins, causing disruptions like genetic disorder. The objective of this study is to develop a computational model that predicts driver genes causing such disruptions in body in the early stages using genomic data, aiming to enhance early diagnosis and intervention. Methods: This study utilized a benchmark genomic dataset, which was processed using feature extraction techniques to identify relevant genetic patterns. Several ensemble classification methods, including XGBoost, Random Forest, LightGBM, ExtraTrees, Bagging, and a stacked ensemble of classifiers, were applied to assess the predictive power of the genomic features. The model, eNSMBL-PRED, was rigorously validated using multiple performance metrics such as accuracy, sensitivity, specificity, and Mathew’s correlation coefficient. Results: The proposed model demonstrated superior performance across various validation techniques. The self-consistency test achieved 100% accuracy, while the independent set and cross-validation tests yielded 96% and 96% accuracy, respectively. These results highlight the model's robustness and reliability in predicting Genetic disorder-related genes. Conclusion: The eNSMBL-PRED model provides a promising tool for the early detection of genetic biomarkers associated with the disorder. In the future, this model has the potential to assist healthcare professionals, particularly doctors, and psychologists, in diagnosing and formulating treatment plans for Genetic Disorder at its earliest stages
  • Thumbnail Image
    Item
    An Intelligent Technical and Vocational Education and Training (TVET) Course Recommendation System based on the Trainee’s Aptitude
    (UMT,Lahore, 2025-01-20) Rana Hammad Hassan
    Personality encompasses the distinct patterns of thoughts, emotions, and behaviors that differentiate individuals and typically remain stable throughout one's life. Aligning these individual traits with learning aptitudes holds promise for improving course outcomes, maximizing returns on investment, and reducing dropout rates significantly. This interdisciplinary research bridges insights from Computer Science (CS) and Human Psychology by analyzing data from Technical and Vocational Education and Training (TVET) programs, focusing on the Big Five Personality traits (BFI). This study marks a pioneering effort in both the Pakistani and global TVET sectors, linking TVET learning skills with individual personalities. We have addressed important ethical considerations, including data privacy and informed consent, ensuring the responsible use of human subjects in this study.