Highlights in Web Engineering II (HIWE II)

Chapter 1

A Review of Generative Models for Facial Expression Recognition

By Ho Jia Jun [0009-0006-6567-8869], Khoh Wee How [0000-0002-7338-8427], Pang Ying Han [0000-0002-3781-6623] and Yap Hui Yen [0000-0002-1367-3226]

Abstract – Despite being widely employed and ubiquitous in many different fields these days, facial expression recognition (FER) still offers significant potential for innovation and broader application. While conventional approaches to FER typically depend on image processing techniques and classical Machine Learning (ML) algorithms, deep learning models are incorporated to enhance performance and produce state-of-the-art outcomes. Feature extraction, model fine-tuning and deep learning models are commonly applied to improve performance and achieve state-of-the-art performance in FER. However, the application of generative models to FER is seldom conducted. In addition to resolving data imbalances, these models have great potential for enriching datasets, generating synthetic facial expressions, and improving robustness against occlusion and noise. In this chapter, a comprehensive study of some common image processing methods, ML and deep learning models, and some possible generative models that can be applied with the learning models in FER will be discussed.

Cite this chapter as:
Ho, J. J., Khoh, W. H., Pang, Y. H., & Yap, H. Y. (2026). A Review of Generative Models for Facial Expression Recognition. In S.-C. Haw, P. Naveen, & J. J (Eds.), Highlights in Web Engineering II (HIWE II) (pp. 3-36). MMU Press.

Chapter 2

Emogram: A Mobile-Based Speech Emotion Recognition Application

By Tan Choon Shen [0009-0007-1326-4039] and Lew Sook Ling [0000-0003-4545-1163]

Abstract – Speech Emotion Recognition (SER) remains a challenging problem in human-computer interaction, particularly on mobile platforms where real-time processing and user privacy are critical. Many existing solutions rely on cloud-based inference, which introduces latency and privacy concerns, or lack robust emotion classification from speech alone. This chapter presents Emogram, a mobile SER application that addresses these limitations through on-device emotion recognition powered by a fine-tuned HuBERT model. The model was trained on the Ryerson Audio-Visual Database of Emotional Speech and Song (RAVDESS) dataset, covering eight emotion classes, and achieved an accuracy of 89% during evaluation. A key innovation of this work is the use of ONNX Runtime to deploy the trained model directly on Android devices, eliminating the need for external servers and enabling offline, low-latency inference. The mobile application was developed using Flutter and integrates Firebase for authentication and data storage, and Cloudinary for media management. Functional testing confirmed the correct operation of all user interface components, including live recording, file upload, prediction history, and emotion trend visualisation. Data management through Firestore and Cloudinary was verified to be consistent and reliable. The results demonstrate the practicality of deploying real-time SER applications on mobile platforms, contributing to the development of accessible tools for emotional health monitoring and intelligent human-computer interaction.

Cite this chapter as:
Tan, C. S., & Lew, S. L. (2026). Emogram: A Mobile -Based Speech Emotion Recognition Application. In S.-C. Haw, P. Naveen, & J. J (Eds.), Highlights in Web Engineering II (HIWE II) (pp. 37-78). MMU Press.

Chapter 3

An ARCS-Based AI Storytelling Application for Motivated Learning in Children

By Wen-Kai Ong [0009-0005-9454-3085], Lee-Yeng Ong [0000-0003-4749-3490] and Meng-Chew Leow [0000-0001-6327-0735]

Abstract – Storytelling is widely recognized as a powerful tool for fostering literacy, creativity and emotional intelligence in children. It plays a vital role in supporting cognitive and socio-emotional development while nurturing a lifelong love for learning. However, traditional storytelling often lacks interactivity and typically requires a parent or teacher, limiting its accessibility and engagement. This determines the need for more dynamic, autonomous and developmentally appropriate forms of storytelling. With rapid advancements in Artificial Intelligence (AI), particularly in Large Language Models (LLMs), it is now possible to deliver personalized, interactive experiences of storytelling with little or no adult intervention. This study proposes the design of a gamified storytelling mobile application that integrates AI technologies such as LLMs, text-to-image, and text-to-speech synthesis. The current work primarily focuses on system design and methodological discussion, while future research will involve conducting large-scale user evaluations with children to quantitatively measure learning motivation and educational outcomes. The application is proposed to help children learn literacy and moral values using interactive, independent stories. The designed application applies to John Keller’s (1987) ARCS (Attention, Relevance, Confidence, Satisfaction) Model of Motivational Learning Design to induce and sustain children’s learning motivation.

Cite this chapter as:
Ong, W.-K., Ong, L.-Y., & Leow, M.-C. (2026). An ARCS-Based AI Storytelling Application for Motivated Learning in Children. In S.-C. Haw, P. Naveen, & J. J (Eds.), Highlights in Web Engineering II (HIWE II) (pp. 79-104). MMU Press.

Chapter 4

Optimising Sales Strategies in E-Commerce Through Data Driven Insight

By Khoo Yeak Rui [0009-0003-6102-0785] and Liew Tze Hui [0000-0001-5069-8441]

Abstract – This study builds predictive models of e-commerce sales trends and analyses the purchasing behaviour of customers using Online Shoppers Purchasing Intention Dataset from Kaggle. The research applies comprehensive data preprocessing including cleaning to handle duplicates and missing values, feature engineering and Synthetic Minority Oversampling Technique (SMOTE) to address class imbalance. Random Forest (RF), Support Vector Classifier (SVC), Decision Tree Classifier (DTC), and Extreme Gradient Boosting (XGBoost) were trained and evaluated to predict purchasing outcomes. XGBoost demonstrated superior performance with 90% Accuracy and AUC score of 0.93 show that driven by behavioural features such as “PageValues”, “Month” and “Operating System” The findings provide critical insights into optimising e-commerce strategies. It is recommended to boost engagement on high-value pages, personalize offers for new visitors, and align inventory with seasonal demand. The results confirm the value of ensemble models in predicting customer behaviour and demonstrate their role in supporting data-driven personalization and strategic decision-making.

Cite this chapter as:
Khoo, Y. R., & Liew, T. H. (2026). Optimising Sales Strategies in E-Commerce Through Data Driven Insight. In S.-C. Haw, P. Naveen, & J. J (Eds.), Highlights in Web Engineering II (HIWE II) (pp. 105-140). MMU Press.

Chapter 5

Customisable OMR Automated Marking System for Efficient Educational Assessments

By Elvin Soon Jung Wong [0009-0003-8983-6508], Yee Jian Chew [0000-0002-2952-986X], Shih Yin Ooi [0000-0002-3024-1011] and Ying Han Pang [0000-0002-3781-6623]

Abstract – Customizable optical mark recognition (OMR) automated marking systems address critical inefficiencies in traditional OMR technologies, which often rely on rigid templates, proprietary hardware, and manual data entry. This chapter presents an open-source solution that leverages computer vision (OpenCV) and Optical Character Recognition (OCR) to automate the grading of multiple-choice answer sheets while accommodating freely designed templates. The system introduces adaptive algorithms for bubble detection, minor alignment correction, and partial mark validation, thereby overcoming the limitations of conventional OMR tools in handling skewed or imperfectly filled sheets. Through OCR integration, it also removes the need to perform manual transcriptions of students’ information, simplifying the assessment process in terms of the work and time taken by educators and institutions. The most important innovations are the template-added platform, which enables users to create their answer keys, bubble patterns, and text boxes (e.g., names/IDs) through a JSON-based configuration; hence, the platform remains flexible to most exam types. These are dynamic thresholding and contour analyses that are used to reliably detect marked responses despite suboptimal scanning conditions. Testing on 95 exam sheets proved a 97.8% grading accuracy, and it performed well when chapter sheets were tilted or sheets were not fully filled. In addition, the OCR module successfully performed readable extraction of handwritten student information in 82 percent of the examinations, but difficulties were encountered in severe cases of irregular handwriting. The proposed system fills a serious market gap in educational technology, as it is an economical yet scalable and comfortable alternative to commercial solutions in the area of OMR. This work highlights that with the removal of administrative overhead associated with large-scale assessments and the reduction of human error, customizable automation of large-scale assessments has the potential to significantly impact academia and many other realms.

Cite this chapter as:
Wong, E. S. J., Chew, Y. J., Ooi, S. Y., & Pang, Y. H. (2026). Customisable OMR Automated Marking System for Efficient Educational Assessments. In S.-C. Haw, P.Naveen, & J. J (Eds.), Highlights in Web Engineering II (HIWE II) (pp. 141-166). MMU Press.

Chapter 6

Design and Development of a Secure and User-Friendly Password Management System with AES-256 Encryption and Two-Factor Authentication

By Praven Kumar Tharmalingam [0009-0004-0792-1854], Yee Jian Chew [0000-0002-2952-986X], Shih Yin Ooi [0000-0002-3024-1011] and Ying Han Pang [0000-0002-3781-6623]

Abstract – Currently, in the digital age Cyberattacks such as data breaches, phishing, and brute-force attacks are extremely common in the digital age, and strong password management has become an absolute necessity. Strong passwords prevent accounts from being compromised, which can be devastating for both individuals and companies. This chapter proposes a secure, robust, and user-friendly password management system designed to efficiently manage digital passwords and credentials. This system aims to solve the password management issue using more intelligent state-of-the-art security implemented in terms of usability and incorporates advanced encryption methods. This includes the AES-256 encryption method to securely store passwords and Zero-Knowledge Architecture so that user data is protected and out if it reaches, even in the case of a server hack. A hashed master password created using bcrypt algorithms and salting is extremely difficult to decrypt in the case of a brute-force attack or using a rainbow table. Sensitive user data, such as Two-Factor Authentication (2FA) keys and user-encrypted passwords, are kept secure in a controlled-access database. The proposed system also includes secure session handling with timeout features that automatically disconnect a user if their session is inactive. Thus, if a user forgets to log out, there is no threat of unauthorized access during a session timeout. The proposed password management system (PassSave) is a robust solution to the increasingly complex issues surrounding digital credential management. By combining robust security measures with a user-centric approach, the system fosters better password practice, protects sensitive user data, and ensures a seamless and reliable experience for all users. The proposed system highlights the potential for innovative technology to enhance cybersecurity and safeguard digital assets in an increasingly interconnected world.

Cite this chapter as:
Tharmalingam, P. K., Chew, Y. J., Ooi, S. Y., & Pang, Y. H. (2026). Design and Development of a Secure and User-Friendly Password Management System with AES-256 Encryption and Two-Factor Authentication. In S.-C. Haw, P. Naveen, & J. J (Eds.), Highlights in Web Engineering II (HIWE II) (pp. 167-192). MMU Press.

Chapter 7

AI-Powered Restaurant Chatbot: Enhancing Dining Experience Through NLP and Automation

By Wong Leong Sheng [0009-0009-4383-0032] and Lew Sook Ling [0000-0003-4545-1163]

Abstract – This chapter presents the design and development of an Artificial Intelligence (AI)-powered restaurant reservation chatbot that aims to modernize and improve the traditional reservation process while simultaneously enhancing customer experience and restaurant operations. Conventional systems for handling bookings are often dependent on phone calls and paper-based logs, which introduce challenges such as delays, errors, limited service availability, and increased administrative burden. These inefficiencies can negatively impact customer satisfaction and are particularly problematic for small-to medium-sized restaurants that may lack the resources to manage reservations effectively. To address these issues, the proposed system enables users to make, modify, and cancel reservations using natural language through a chatbot interface, ensuring more intuitive and user-friendly interaction. Beyond reservation handling, the system incorporates additional features, including menu browsing, food recommendations generated through a personality quiz, and collection of customer feedback to further enrich the dining experience. The chatbot was developed as a web application using PHP, JavaScript, and MySQL with real-time backend database integration to guarantee data accuracy, consistency, and secure interactions. An admin dashboard was also implemented to allow restaurant staff to manage menu items, bookings, and user accounts efficiently. The development process followed the agile methodology and underwent user acceptance testing, achieving strong success rates across all functional modules. By combining AI-driven conversational capabilities with immediate data processing, the system overcomes the fundamental limitations of traditional reservation approaches, providing a cost-effective, scalable, and intelligent solution ideally suited for small- and medium-sized restaurants seeking to enhance service levels, operational efficiency, and personalization in customer engagement.

Cite this chapter as:
Wong, L. S., & Lew, S. L. (2026). AI-Powered Restaurant Chatbot: Enhancing Dining Experience Through NLP and Automation. In S.-C. Haw, P. Naveen, & J. J (Eds.), Highlights in Web Engineering II (HIWE II) (pp. 193-218). MMU Press.

Chapter 8

A Privacy-Preserving Web-Based Email Analysis and Investigation Tool with Partial Homomorphic Encryption

By Abdullah Omar Hamad Bin Afeef [0009-0007-5337-024X] and Siti Husna Abdul Rahman [0009-0002-5750-6363]

Abstract – Email remains one of the most common communication channels in daily life, yet it is a significant entry point for cyberattacks. Most available email security tools must be given complete access to unencrypted messages, raising serious privacy concerns. This chapter proposes an alternative: web-based email analysis utility that preserves user privacy through Partial Homomorphic Encryption (PHE). The system (called SaveInbox) encrypts sensitive fields (sender, recipient, and subject) while still allowing certain types of threat analysis tasks to run on non-encrypted or partially encrypted data (basic malware scanning or phishing detection on message content). It uses a hybrid encryption approach: the bulk of each email is encrypted with AES-256, and the small AES key and necessary metadata are encrypted using the Paillier PHE scheme. The encrypted data are then analysed third-party security application programming interfaces (APIs) without revealing sensitive information. We developed a prototype and tested it against phishing-, spam-, and malware-laden emails. The results indicate little to no performance penalty and no loss of detection accuracy, which is a promising trade-off between privacy and utility. Altogether, this method lays the groundwork for privacy-respecting email investigations, enabling organizations to leverage external email analysis services without disclosing confidential information.

Cite this chapter as:
Afeef, A. O. H., & Abdul Rahman, S. H. (2026). A Privacy-Preserving Web-Based Email Analysis and Investigation Tool with Partial Homomorphic Encryption. In S.-C. Haw, P. Naveen, & J. J (Eds.), Highlights in Web Engineering II (HIWE II) (pp.219-240). MMU Press.

Chapter 9

GrocGuardian: A Smart Web-Based Assistant for Grocery Expiry Tracking and Ingredient-Aware Recipe Suggestions

By Jayapradha J [0000-0002-2548-9135], Su-Cheng Haw [0000-0002-7190-0837], Rudranil Patra [0009-0008-4179-1343] and Senthil Kumar T [0000-0002-2200-3339]

Abstract – This study presents GrocGuardian, a smart, end-to-end web-based aid that maximizes household grocery management in response to the increasingly growing challenges of global food waste and increasing demands for environmentally conscious household practices. The application is found on three essential functionalities: tracking expiry dates, reminder notifications in advance, and smart recipe suggestions from the available pantry items. By embracing cutting-edge technology stacks, such as React, TypeScript, Node.js, and MongoDB/PostgreSQL, GrocGuardian provides a responsive and modular user interface. The solution provides real-time state management, end-to-end secure user authentication, and scalable features that are appropriate for integration with future IoT devices and AI-based personalization. Unlike current siloed options, GrocGuardian provides an end-to-end solution for reducing food waste and encouraging efficient and sustainable consumption. It prescribes the system architecture, implementation plan, and performance metrics of GrocGuardian, and sets it up as a building block for future smart kitchen ecosystems.

Cite this chapter as:
Jayapradha, J., Haw, S.-C., Patra, R., & Senthil Kumar, T. (2026). GrocGuardian: A Smart Web-Based Assistant for Grocery Expiry Tracking and Ingredient-Aware Recipe Suggestions. In In S.-C. Haw, P. Naveen, & J. J (Eds.), Highlights in Web Engineering II (HIWE II) (pp. 241-258). MMU Press.

Chapter 10

Machine Learning Approaches for Forecasting Oil, Gas, and Water Production in Reservoir Engineering

By Keat Fu Tham 0009-0007-2757-6877, Rajasvaran Logeswaran [0000-0001-5784-486X], Sellappan Palaniappan [0009-0009-1168-2864] and Harvin Kaur Gurchran Singh [0000-0003-3758-6677]

Abstract – Accurate production forecast is of the utmost importance in upstream oil and gas. Accurate forecast can ascertain remaining oil recoverable reserves, enable profitable execution of infill drilling, well workovers, and stimulation projects and proper reservoir management in prolonging field, and asset life. Numerical reservoir simulation provides robust and accurate production forecast with proper history matching and accurate static model but is time and effort consuming. They are also computational expensive and subsurface uncertainties are high. Decline Curve Analysis (DCA), an empirical type curves method, is used extensively due to its simplicity and fast turnaround in providing robust forecast. However, both methods require deep understanding of the reservoir behaviours and underlying physics. The advent of machine learning algorithms has provided an alternative for production forecast leveraging on readily abundant and available, production and subsurface data. This chapter aims to investigate the usability of machine learning algorithms in hydrocarbon production forecas.

Cite this chapter as:
Tham, K. F., Logeswaran, R., Palaniappan, S., & Singh, H. K. G. (2026). Machine Learning Approaches for Forecasting Oil, Gas, and Water Production in Reservoir Engineering. In S.-C. Haw, P. Naveen, & J. J (Eds.), Highlights in Web Engineering II (HIWE II) (pp. 259-298). MMU Press.

Chapter 11

Optimising Diabetes Prediction Using GAN-Based Approach and Explainable AI

By Alicia Chua Yang Qin [0009-0008-8015-7644], Maizatul Akmar Binti Ismail [0000-0003-1877-7128], Riyaz Ahamed Ariyaluran Habeeb [0000-0002-6064-1016]  and Jamallah M. Zawia [0009-0003-7767-5794]

Abstract – Diabetes today is a significant health issue worldwide as it tends to cause some severe complications such as diabetic retinopathy and kidney failure, thus the need to detect it at an early stage. We address this problem by developing a framework that combines Conditional Tabular Generative Adversarial Networks (CTGAN) with SHapley Additive exPlanations (SHAP) to increase the accuracy and understandability of predictions of diabetes. CTGAN resolves the issue of unequal classes by creating realistic artificial samples in the minority class to lessen bias in machine learning models. SHAP offers an understanding of the behavior of the model to make predictions, and this features that are important. We used a subset of CDC BRFSS 2021 as a test and trained machine learning models to predict diabetes, and CTGAN achieved better results than such methods as Synthetic Minority Oversampling Technique (SMOTE). XGBoost with CTGAN, our best model, increased the F1 score by approximately 40 with respect to SMOTE-based approaches, which gave an accuracy of 85.30%. In contrast to SMOTE that produced less realistic data, the CTGAN realized a precision and a recall of 70-80. The analysis of the SHAP identified two factors as the strongest in promoting diabetes risk Body Mass Index (BMI) and age. Combining CTGAN and SHAP provides clinicians with easy to understand and an actionable information on how to detect diabetes at an early stage.

Cite this chapter as:
Chua, A.Y.Q., Ismail, M.A., Habeeb, R.A.A., & Zawia, J.M. (2026). Optimising Diabetes Prediction Using GAN-Based Approach and Explainable AI. In S.-C. Haw, P. Naveen, & J. J (Eds.), Highlights in Web Engineering II (HIWE II) (pp. 299-340). MMU Press.