UTAR Institutional Repository (Universiti Tunku Abdul Rahman)
Not a member yet
6132 research outputs found
Sort by
Predictive analytics student dropout rate and academic success rate
Get PDFThe student dropout problem is still a critical problem that occurred at many universities, that influencing both institutional reputation and student academic success. The project objective is to design and develop an interactive dashboard that can predict and visualize the student academic success and dropout patterns using Power BI integrated with a trained model which is LightGBM. The dashboard would be applicable to both educators and administrators with a real-time, user-friendly, and personalized insights that support data-driven intervention strategies for at-risk students. Before the development of dashboard, the CRISP-DM process would be started like data collection, preprocessing, modelling, and visualization of model. The dataset is obtained from a Portuguese university was used, collaborating with demographic, financial, and academic data. Data preparation have including handling missing values, perform sampling methods for data. LightGBM has been selected for deployed model in Power BI because it is good for handling multi-class and during handling high-dimensional datasets always can result in good accuracy rate and intpretabiity. The dashboard have contained with seven modules which are General Overview, Demographic Analysis, Financial Analysis, Model Performance Analysis, StudentRISK Navigator, Navigation Guide and Dataset Descriptions. These modules had provided professional insights and predictive accuracy that can reduce the gap between predictive models and user decision making in education domain. It can allowed institutions to take personalized interventions to help the at-risk students that can improve their retention rate and academic success rate
Deep learning: diabetic retinopathy detection using fundus image
Get PDFDiabetic retinopathy (DR) is one of the leading causes of blindness worldwide. DR
normally come with diabetes when it affects an individual. Since DR is one of the
leading causes that cause blindness, early detection of it is much essential to prevent
vision loss. However, the traditional method of diagnosing DR is to rely on manual
examination of retinal images by ophthalmologists. It is a time-consuming and is
highly prone to error process since it is done by using manpower. Nowadays, with the
growing number of diabetic patients, there is an urgent need for an efficient and
automated solution for DR screening to prevent blindness due to DR, and therefore
this project is proposed. This project is aimed to develop a deep learning-based
system using Convolutional Neural Networks (CNNs) for the automated detection and
classification of DR. The dataset that is applied in this project is obtained from
Kaggle “Diabetic Retinopathy 224 x 224 Gaussian-Filtered” dataset, which consist
high-resolution fundus images across the five classes: No DR, Mild, Moderate, Severe
and Proliferate DR. Due to the significant class imbalance appear in the dataset, some
data augmentation techniques such as flipping, rotation, and zooming, along with
class weighting are applied to improve the performance. A DenseNet-201 model with
transfer learning from ImageNet was employed, enhanced with global average
pooling, batch normalization, dropout, and a softmax output layer. Some metrics were
used to evaluate the performance of the model such as accuracy, precision, recall, F1-
score and confusion matrix analysis. The experiment results show that the proposed
DenseNet-201 model achieves strong classification performance and shows promise
as a reliable and efficient tool for automated DR screening, supporting early detection
and reducing the workload of the ophthalmologists
Token processing in digital asset transaction platform
Get PDFThis project has developed FC Uniswap—a comprehensive decentralised exchange platform designed to address critical challenges in digital asset token processing. The system integrates the Uniswap V2/V3 protocol with LayerZero V2 cross-chain bridging technology, demonstrating advanced token handling capabilities within digital asset trading platforms. Implementation focused on three core objectives: enhancing interoperability between blockchain platforms through a cross-chain communication framework; optimizing coordination in asset tokenization processes via a Byzantine fault-tolerant synchronization protocol; and strengthening smart contract security through a hybrid verification system.
The project employs a layered architecture: the frontend is built using Next.js, smart contract development utilises the Hardhat tooling, and blockchain interactions leverage the ethers.js v6 library. Core technological innovations include a burn-and-mint cross-chain bridge mechanism that enables asset transfers between Polygon Amoy and Ethereum Sepolia testnets, multi-compiler support compatible with Solidity versions 0.4.19 to 0.8.20, and the integration of Uniswap V3's concentrated liquidity feature to enhance capital efficiency.
Testing demonstrated a 100% success rate in core functionality (all 14 test cases passed) and an over 95% transaction success rate in mainnet fork environments. The project successfully resolves interoperability challenges through standardised cross-chain protocols, streamlines development workflows to reduce process fragmentation, and implements foundational security measures, including access controls and re-entrancy protection
Virtual classroom platform with real-time collaboration features
Get PDFThis project situated in the field of educational technology, develops a web-based Virtual Classroom Platform to enhance online learning through real-time collaboration. Current virtual learning environments often suffer from inadequate real-time interaction, poor accessibility for users with limited technological resources, and ineffective peer-to-peer engagement, collectively diminishing the learning experience. To address these, the platform employs frontend technologies (HTML, CSS, JavaScript) for a responsive and device-agnostic interface, backend technologies (PHP, MySQL) for robust data management, and Node.js server using Socket.IO for real-time communication. It integrates open-source collaboration tools such as Jitsi Meet for video conferencing, Etherpad for collaborative text editing, and WBO for shared whiteboards to enable synchronous learning and foster student-instructor peer interactions. The development process uses the Agile methodology, which emphasizes flexibility and iterative progress. Design and functionality were informed by user requirements gathered through a questionnaire, guaranteeing that the platform satisfies a range of needs. The resulting platform provides a dynamic, inclusive, and interactive environment that reduces the distinction between online and in-person education. It overcomes existing limitations by ensuring accessibility across devices and technological capabilities, with successful integration of collaboration tools enhancing user experience. This project excludes the development of mobile applications, automated grading, intelligent tutoring, or personalized learning paths, and VR/AR technologies, focusing on traditional web-based tools for real-time communication and collaboration. This work advances online education by providing a scalable, equitable solution that supports real-time interaction and collaboration, contributing to a more effective learning ecosystem
Mandarin learning app for english-speaking children
Get PDFMandarin has become one of the most important languages in the world due to China's significant influence on economics, politics, and culture. Although its economic and political rise has increased its importance, learning Mandarin as a second or foreign language remains a considerable challenge, particularly for native English-speaking children due to tonal variations, complex grammar, and limited access to engaging and age-appropriate resources. However, most existing learning applications, such as Duolingo, HelloChinese, Studycat, and ChineseSkill, provide learning materials that teach Mandarin via English but lack essential features, including real-time translation, text or image summarization, and personalized learning paths. These applications often prioritize vocabulary over critical language skills, such as pronunciation and grammar, while adopting a one-size-fits-all approach that does not adapt to individual proficiency levels. This project aims to deliver a user-friendly mobile application that helps English-speaking children aged 10 to 12 effectively learn Mandarin through English as an assisting language, addressing limitations in existing applications. The proposed application provides interactive lessons and exercises that cover Pinyin, Chinese characters, grammar, and speaking skills. Advanced features such as real-time translation and a summarization tool are integrated into the proposed application for self-directed learning. It also includes interactive elements, such as real-time feedback and a reward system, to motivate the children throughout the learning process. To ensure personalization, the system includes an entry-level test and leverages a Support Vector Machine model to predict learner performance. This model achieved 97% accuracy and 95.65% F1-score, allowing the application to recommend target review sessions for learners predicted to perform poorly. The proposed application is developed using agile methodology and technologies such as Android Studio, Amazon Web Services, Firebase, Jupiter Notebook, the Gemini API, FastAPI and Google Translation packages
Optimisation of extraction from Tradescantia zebrina leaves and the gastrointestinal stability of its phytochemicals and bioactivities
Get PDFTradescantia zebrina is a medicinal plant traditionally consumed by residents worldwide, which is extensively investigated for its phytochemical contents and bioactivities. However, the effects of gastrointestinal (GI) digestion on the potential health benefits are poorly understood. This study was conducted to optimise the extraction of phytochemical from T. zebrina leaves. The optimal extract was subjected to simulated GI digestion based on the INFOGEST 2.0 protocol. The current study discovered that sequential hybrid extractions were more effective than the individual methods like hot water extraction (HWE) and ultrasonic-assisted extraction (UAE). The “UAE for 20 min, followed by HWE for 15 min” extract was the optimal among all nine treatments, showing the highest total phenolic content (TPC) (8.114 ± 0.007 mg GAE per g dry extract) and total flavonoid content (TFC) (62.56 ± 0.29 mg QE per g dry extract). This extract also recorded the lowest EC50 in 1,1-diphenyl-2-picrylhydrazyl radical (DPPH•) (0.587 ± 0.000 mg/mL) and 2,2′-azino-bis(3-ethylbenzothiazoline-6-sulphonic acid) (ABTS•+) (1.069 ± 0.003 mg/mL) scavenging activities. Simulated GI digestion significantly reduced TPC and TFC, respectively, to 3.277 ± 0.005 mg GAE and 6.44 ± 0.22 mg QE per g dry extract. Simulated GI digestion also declined the antioxidant and anti-inflammatory activities of the selected T. zebrina extract. The EC50 values of DPPH• and ABTS•+ scavenging activity increased to 2.118 ± 0.042 and 1.585 ± 0.001 mg/mL, respectively. The EC50 values of hydrogen peroxide scavenging activity and inhibition of albumin denaturation increased, respectively, from 1.976 ± 0.006 to 3.658 ± 0.023 mg/mL and from 1.527 ± 0.005 to 2.918 ± 0.023 mg/mL. These changes could potentially proof the negative impacts of human GI digestion, signalling the necessity for strategies that enhance the bioaccessibility
The predictive roles of social anxiety, stress, and boredom proneness on smartphone addiction among university students in Malaysia
Get PDFIn Malaysia, smartphone addiction has become a significant issue, with the country ranking third worldwide in prevalence. University students are especially vulnerable, given their extensive use of smartphones for academic, social, and personal purposes. Excessive smartphone reliance has been linked to psychological factors such as social anxiety, stress, and boredom proneness. However, limited research has explored how these variables contribute to smartphone addiction within the Malaysian context. Thus, the present study aimed to explore the predictive roles of social anxiety, stress, and boredom proneness on smartphone addiction among Malaysian university students, guided by the Compensatory Internet Use Theory (CIUT). A quantitative, cross-sectional design was employed, and data were collected through an online survey. A total of 113 participants were recruited using purposive and snowball sampling techniques. The inclusion criteria required participants to be (i) Malaysian undergraduates, (ii) aged between 18 and above, and (iii) smartphone users. It was hypothesised that social anxiety, stress, and boredom proneness would positively predict smartphone addiction. The findings revealed that social anxiety and boredom proneness positively predicted smartphone addiction, whereas stress was not a significant predictor. These findings extend the application of CIUT by highlighting its relevance to social anxiety and boredom proneness, but its limitations in explaining stress. The study contributes to the literature on digital well-being and provides practical implications for educators, policymakers, and mental health professionals to design interventions that reduce overreliance on smartphones through addressing underlying psychological factors
Species distribution model to predict the occurrence of Malayan partridge
Get PDFClimate change has caused several problems in Malaysia such as increase of temperature and change in precipitation patterns. Malayan Partridge (Arborophila campbelli) is a bird species found in Peninsular Malaysia that is facing the threat of habitat loss due to climate changes. Currently, this species is understudied and that leads to less information about the future occurrence of this species. Therefore, this study aims to produce a prediction of current and future occurrence of Malayan Partridge in Peninsular Malaysia with different models Species Distribution Model (SDM) that are Maximum Entropy Model (MaxEnt), Random Forest (RF), Support Vector Machine (SVM), Generalized Linear Model (GLM) and Bioclim. Different pseudoabsence data settings will be implemented to identify the best setting to predict the occurrence of the species. Species occurrence data were collected from public biodiversity databases 19 bioclimatic variables were sourced from WorldClim to predict the current occurrence of the species. A variable selection process will be used to identify the important bioclimatic variables. These variables will be used for the models of SDM. To predict the potential future occurrence of the species, Shared Socioeconomic Pathway (SSP) will be implemented. The performance of the model will be evaluated through Area Under Curve (AUC) and cross-validation techniques. Habitat suitability maps will be produced because of the model to provide visualization
Traffic signal control with deep reinforcement learning
Get PDFTraffic congestion, one of the problems that impact a large population of mankind, has
incurred loss in terms of time, fuel, money and pollution. The solution lies in traffic
signal control (TSC), optimized control is the way to mitigate traffic congestion. With
numerous efforts and research in this area, the methods have evolved from
transportation theory to deep reinforcement learning (DRL) approaches over the years.
This report presents extensive research on various direction in this domain and
identifies the gap of previous research and real-world deployment. The project restudies
the nature of the problem, and therefore, propose a new formulation of Markov decision
process (MDP) and framework in TSC to improve efficiency and generalizability of the
algorithm in various scenario. Furthermore, this project explores the improvement of
Soft Actor Critic (SAC) with gradient-based meta learning (GBML) method.
Comprehensive experiments are conducted on Simulation of Urban Mobility (SUMO)
to evaluate the effectiveness of the algorithm
Application for cow lameness detection using computer vision
Get PDFLameness is a major welfare and economic issue in dairy farming, causing reduced milk yield,
reproductive failure, and premature culling. Small and medium-sized farms often lack
affordable tools for early detection, leading to delayed treatment. This project develops a cost
effective mobile application that uses computer vision to detect lameness in cows and provides
basic herd management functions. The lameness detection model applies YOLOv8-based pose
estimation to extract anatomical key points and calculate back arching through Root Mean
Squared Error (RMSE), identifying abnormal postures linked to lameness. A cow recognition
system, based on ResNet18 and ArcFace embeddings, allows farmers to register cows by coat
patterns and maintain health records. The system is supported by a FastAPI backend with
Firebase integration for storage, authentication, and data management. Through a simple
Flutter-based mobile app interface, farmers can upload images, receive predictions, confirm
cow identities, and update records. This approach enables early diagnosis, improves animal
welfare, and reduces economic losses