Theses
Permanent URI for this communityhttps://rda.sliit.lk/handle/123456789/2429
Postgraduate students are required to submit a thesis as part of fulfilling the requirements of their respective postgraduate degree programmes. This community features merit-based graduate theses submitted by SLIIT postgraduate students. Abstracts are available for public viewing, while the full texts can be accessed on-site within the library.
Theses and Dissertations of the Sri Lanka Institute of Information Technology (SLIIT) are licensed under a
Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License
.
Browse
11 results
Search Results
Publication Open Access Development of an AI-Integrated Online Counseling and Self-Improvement Platform for Mental Health Support(Sri Lanka Institute of Information Technology, 2026-01) Wijewardena T PMental health challenges such as stress and anxiety remain a growing global concern, particularly in low-resource settings like Sri Lanka, where access to professional counseling is limited and stigma discourages many individuals from seeking support. This thesis presents the development of an offline AI-based counseling chatbot designed to provide accessible, empathetic, and private mental health support without relying on high-bandwidth internet connections. The system was implemented using a TF-IDF-based natural language processing pipeline to classify user inputs into predefined intent categories and deliver evidence-based therapeutic responses. Training data were compiled from clinical counseling transcripts, standardized affective word databases, anonymized peer support forums, and publicly available datasets, ensuring both linguistic diversity and clinical relevance. Evaluation of the system demonstrated an intent detection accuracy of 91.2% across 387 test queries. A preliminary user study involving 10 participants revealed that 80% reported noticeable stress reduction after interaction, while responses were rated at an average of 4.3/5 for relevance. The chatbot maintained a lightweight design with an average response time of 0.19 seconds and a memory footprint of just 2MB, enabling reliable operation on low-end devices in offline settings. The findings confirm that simple, transparent AI techniques can effectively bridge treatment gaps in underserved regions. While the current system is English-only, future enhancements will focus on incorporating multilingual support, contextual emotion analysis, and improved personalization, providing a scalable and culturally adaptive solution for equitable mental health care.Publication Open Access AI-DRIVEN SELF-HEALING TEST AUTOMATION FOR ENTERPRISE SOFTWARE SYSTEMS(Sri Lanka Institute of Information Technology, 2025) Jinarathna, H. D. R. J.Enterprise Resource Planning (ERP) and Customer Relationship Management (CRM) platforms are backbone tools for today’s businesses, helping teams around the company work together efficiently. But because these systems are huge and always changing, testing them gets tricky. Methods we’ve traditionally used—whether having testers run scripts by hand or running automated scripts—can’t keep up. To solve these problems, I’ve built a test automation framework that uses AI to repair itself as it runs. By incorporating Natural Language Processing (NLP) for spotting system changes and Reinforcement Learning (RL) for teaching tests to heal and get sharper, the framework learns to find when a test has broken, to fix the code on the fly, and to keep fine-tuning itself, so people hardly need to step in. I shaped the system by talking to QA engineers about the roadblocks and running pilot cases in a pretend ERP setup. Those conversations, plus the numbers, helped us tweak the design so it feels less like a lab gadget and more like a teammate. Early results are encouraging—flake tests bounce back 35% more often and testers spend 25% less time rewriting logic by hand. My research helps Software Quality Assurance Engineers, learners, and software businesses by offering an easy-to-understand, adaptable way to test big-company software. The results show how using smart technology can make software testing faster, cheaper, and better.Publication Open Access AI-Driven Help Desk Integration: Enhancing Customer Support with Chatbots, Sentiment Analysis, and SLA Automation(Sri Lanka Institute of Information Technology, 2025-11) Nimnadi DilsikaThis research investigates the integration of Artificial Intelligence (AI) into help desk systems to improve customer service efficiency, accuracy, and overall satisfaction. The proposed AI-driven help desk framework combines three intelligent components: a chatbot for instant and automated responses, a sentiment analysis engine to detect and interpret customer emotions, and a Service Level Agreement (SLA) management module that ensures real-time tracking of response and resolution performance. Using a dataset of 40,000 simulated support tickets, the system was evaluated for key metrics such as response time, SLA compliance, and customer satisfaction levels. The results demonstrated notable improvements, including faster response rates, higher SLA adherence, and enhanced emotional understanding in customer interactions. Overall, the study confirms that AI integration transforms traditional help desks into proactive, data-driven, and emotionally intelligent service environments. Future advancements will focus on predictive SLA modeling, multilingual capabilities, and multimodal sentiment analysis for broader adaptability.Publication Open Access Development of a Non-Invasive Algorithm for Anemia Detection in Women in Sri Lanka(SLIIT, 2024-12) Senanayake, W.I. UmayaAnemia continues to be a considerable health issue for women in Sri Lanka, impacting physical and cognitive growth, general health, and economic efficiency. Diagnostic methods, like blood tests, are invasive, time-consuming, and could be out of reach for populations with limited resources. A non-invasive algorithm is created to detect anemia in Sri Lankan women in this thesis. The algorithm utilizes readily available clinical and demographic information to decrease reliance on conventional blood tests. According to that ―Development of a Non-Invasive Algorithm for Anemia Detection in Women in Sri Lanka‖ entitled as the research title of this thesis. The research involves data collection from women across varied demographics and regions, combined with vital health parameters and physical indicators relevant to anemia detection. Advanced machine learning models are trained on this data to identify patterns associated with anemia, offering accurate predictions without the need for invasive procedures. A core aim of the study is to enhance early detection, enabling timely intervention and reducing the overall prevalence of anemia among women. The high sensitivity rate of the algorithm allows for effective anemia detection with minimal input data, according to key findings. Furthermore, its non-invasive characteristics make it appropriate for application in rural regions where healthcare resources are scarce. The system successfully provides a non-invasive, accurate, and accessible method for anemia detection, using fingertip imaging and machine learning to predict anemia in real-time. With a compact device integrated into a web app, users can monitor their health easily, while healthcare providers can remotely access patient data for timely interventions. The system’s cost-effectiveness and ease of use make it particularly valuable for resource-limited settings, offering a scalable solution for anemia management and broader public health impact.Publication Open Access Development and Integration of an AI-Driven PHP Adapter for Automated Mathematical Question Classification and Assessment: Enhancing Student Profiling and Feedback Mechanisms(SLIIT, 2024-12) Nishamali, M.K.C.P.The transformative growth of AI can be seen in almost every sector. AI can be a useful application for the educational domain as well. This research aims to combine IT to develop mathematics subjects by leveraging AI in practice mainly introducing Capabilities of Open AI. The primary objective is to create OpenAI API through a specially created PHP adapter to classify mathematical questions into six main themes Sets and Probability, Algebra, Numbers, Geometry, Measurements, and Statistics. This automated AI-driven classification system helps to create online assessments within the blink of an eye. The Integration of Open AI API with a PHP-based framework makes a bridge between AI capabilities and education needs. This framework is the ideal solution for manual and traditional school assessments. This plugin can be implemented in other university-level courses as well. The sample of the adapter plugin is only created and tested for secondary school mathematics classes for grade 10. This AI-driven mathematics classification system is designed to optimize the assessment process by providing additional objectives such as leveraging automated student grading feedback so teachers and students can see the result instantly. Additionally, answers are automatically generated after the assessment, displaying the solving steps that help students identify their mistakes. Meanwhile, this system also predicts the student’s mathematics pass mark based on the results of the tests taken from this system.Publication Open Access Development of a Non-Invasive Algorithm for Anemia Detection in Women in Sri Lanka(SLIIT, 2024-12) Senanayake, W.I.UAnemia continues to be a considerable health issue for women in Sri Lanka, impacting physical and cognitive growth, general health, and economic efficiency. Diagnostic methods, like blood tests, are invasive, time-consuming, and could be out of reach for populations with limited resources. A non-invasive algorithm is created to detect anemia in Sri Lankan women in this thesis. The algorithm utilizes readily available clinical and demographic information to decrease reliance on conventional blood tests. According to that ―Development of a Non-Invasive Algorithm for Anemia Detection in Women in Sri Lanka‖ entitled as the research title of this thesis. The research involves data collection from women across varied demographics and regions, combined with vital health parameters and physical indicators relevant to anemia detection. Advanced machine learning models are trained on this data to identify patterns associated with anemia, offering accurate predictions without the need for invasive procedures. A core aim of the study is to enhance early detection, enabling timely intervention and reducing the overall prevalence of anemia among women. The high sensitivity rate of the algorithm allows for effective anemia detection with minimal input data, according to key findings. Furthermore, its non-invasive characteristics make it appropriate for application in rural regions where healthcare resources are scarce. The system successfully provides a non-invasive, accurate, and accessible method for anemia detection, using fingertip imaging and machine learning to predict anemia in real-time. With a compact device integrated into a web app, users can monitor their health easily, while healthcare providers can remotely access patient data for timely interventions. The system’s cost-effectiveness and ease of use make it particularly valuable for resource-limited settings, offering a scalable solution for anemia management and broader public health impact.Publication Open Access Development of an Integrated IoT System for Remote Monitoring and Enhanced Safety Assurance in Outdoor Environments(SLIIT, 2024-12) Dawlagala, D. S. D. M. D.Making places safer outside for freely operating has also risen in priority especially in places with little or no communication facilities and people can go out for camping and hiking. In this study, a distributed Internet of Things (IoT) system incorporating geofencing, environmental monitoring and real-time positioning system to improve outdoor navigation and safety has been developed and tested. The system includes one main hub and a number of sub devices connected to ESP32 microcontroller and LoRa 433 MHz for communication in addition to GPS (Neo-7M) for position location purposes. Linear and angular displacement measurements and their data processing and recording were performed using sub-devices that incorporated GPS and other modules used for receiving and transmitting data over a LoRa network. Sub and the main devices demonstrated data on 0.96 inches OLED both devices which enable users to receive up to date responses. There is also a type of geofencing action in the system, where an alert is sent when a sub-device departs an area that has been defined. The geofencing alerts can be managed though a web dashboard that utilizes Node.js, Next.js, and Web-Sockets for dashboard and main hub interaction. The main hub was also able to send and receive updates from the internet and then transmit this information to the sub-devices over the LoRa network thus bridging the gap that existed between local and remote operation. From the initial results, the newly proposed system is able to provide secure communication, precise location coordinates and prompt geofencing alerts even in outdoor environments with poor network support. The subsequent steps will seek to enhance the energy consumption of the system, improve communication coverage, and make more tests in real-life scenarios to assess the system as a whole.Publication Open Access Development of a Neural Network-Based Framework for Skin Disease Recognition(SLIIT, 2024-12) Senadhipathi, L.A.N.MSkin diseases impact humans, animals, and plants and are typically brought on by germs or infections. These ailments include ringworm, yeast infections, brown sports, allergies, and other conditions. Early detection can help lessen the impact of diseases. But there are other risks that the skin can encounter, one of which is illness. Fungi, bacteria, allergens, enzymes, and viruses are the main causes of skin problems. Skin conditions impair not just one's physical health but also their psychological well-being, especially in those who have damaged or even scarred skin. Identifying the condition via manual feature extractions or symptom-based approaches requires time and requires comprehensive data for accurate identification. Serious health concerns are associated with skin diseases, which require an accurate and timely diagnosis for appropriate treatment. In particular, convolutional neural networks (CNNs) have shown promising results in automated skin disease identification recently. In this study, A novel CNN-based approach is presented, achieving a 95% accuracy rate in classifying seven different types of skin diseases from the HAM10000 image dataset. Dermatoscopic images from the HAM10000 dataset are preprocessed and categorized into seven classes: basal cell carcinoma, melanoma, vascular lesions, dermatofibroma, melanocytic nevi, and benign keratosis. After extensive testing and fine-tuning, it achieved an overall accuracy of 95% on the testing set. The outcomes show that the suggested CNN-based method can accurately identify a variety of skin conditions by using the HAM10000 picture dataset. Deep learning techniques can significantly help dermatologists and other healthcare professionals diagnose skin conditions accurately and automatically, enabling them to provide prompt and efficient treatments. This work adds a great deal to the growing field of dermatological computer-aided diagnosis and offers valuable data for upcoming advancements in the identification of skin diseases.Publication Open Access Development of Queue Estimation Algorithms for Urban Intersections in Mixed Traffic Conditions(Department of Civil Engineering Sri Lanka Institute of Information Technology, 2023-12) Jayatilleke, J.A. D.S.STraffic congestion has increased globally due to rapid urbanization and expedited economic developments in many countries. Vehicle queues are a governing aspect of traffic congestion, studied over the past decades. Most of the existing queue estimation approaches are limited to homogeneous traffic conditions. However, the traffic conditions in many developing countries are heterogeneous and are heavily influenced by mixed vehicle composition, lane changing, and gapfilling behaviors. This study aims to estimate the queue length at signalized intersections having heterogeneous traffic conditions. The methodology employed in this study integrates both statistical and neural network analyses utilizing a time-series approach. A key innovation in this research lies in the incorporation of heterogeneity considerations, where Passenger Car Units (PCU) are assimilated into the measurements of traffic flow and lane-changing movements within the analyzed road section. The influential factors impacting queue length were examined, encompassing arrival flow, discharge flow, outbound lane change, inbound lane change, and signal configuration. The statistical analysis was undertaken through an econometric approach, representing another novel contribution to queue estimation studies. Vector Auto Regression (VAR) models were developed to estimate queue lengths for signalized and unsignalized intersections. The VAR estimation results demonstrated heightened accuracy in queue estimation and practical applicability for prediction, capturing the traffic characteristics of the formed vehicle queue. However, limitations were identified, particularly in terms of lower prediction times, which impeded the practical utilization of the model for traffic management. Consequently, to address this limitation, neural network analysis using the Long Short-Term Method (LSTM) was incorporated to enhance queue predictions over longer time sequences. While the neural network exhibited promise, challenges in data collection contributed to lower accuracy in predictions. Notwithstanding the challenges, the methodological development in this thesis presents a promising direction for queue estimations under heterogeneous conditions. This advancement brings the scientific and research field one step closer to improved queue estimation methods within this specific scope.Publication Open Access Development of Real-Time, Self-Learning Artificial Intelligence-Based Algorithms for Non-Intrusive Energy Disaggregation in a Multi-Appliance Environment(Faculty of Engineering Sri Lanka Institute of Information Technology, 2023-12) Herath, MElectricity serves as a cornerstone in modern economies, with demand in residential and commercial sectors rapidly increasing in recent years. Enabling real-time monitoring of individual appliance-wise energy consumption and delivering user feedback is essential for future energy conservation initiatives. Energy disaggregation becomes imperative in furnishing consumption statistics for individual appliances. The acquisition of appliance-specific energy consumption in a non-intrusive manner, without the need for sensors on each device but by utilizing readings from the main household energy meter, highlights Non-Intrusive Load Monitoring (NILM) as a promising solution. NILM, leveraging the capabilities of smart meters and advancements in computational power, gains popularity for its effectiveness in disaggregating and analyzing energy consumption patterns. This study introduces an Artificial Intelligence (AI)-based NILM solution capable of disaggregating the energy consumption of multiple appliances while adapting to new appliances and their evolving behaviors. Among various NILM approaches, Neural Network (NN)-based models demonstrate promising disaggregation capabilities. However, the selection of the most suitable NN type or architecture poses a challenge due to the multitude of approaches in literature. To address this issue, the study standardizes and compares different NNs, with results showing that the Convolutional Neural Network (CNN) exhibits superior prediction accuracy and speed. This study also investigates the impact of different appliances and their consumption profiles on disaggregation performance, rigorously testing parameters such as NN architecture, input-output mapping topologies, data preprocessing, and hyperparameters. This leads to the development of guidelines for future NILM studies. Additionally, the study introduces a hierarchical plug-and-play modular-based model for appliance anomaly detection, extending the application of NILM and overcoming limitations in anomaly detection literature. This study investigates two-dimensional (2D) input-based NILM solutions for predicting appliance energy consumption profiles and classifying appliances. Unlike conventional NN-based models using 1D signals, representing the aggregate energy signal as a 2D image improves performance by leveraging feature extraction capabilities of NNs and preserving vital temporal information and signal amplitude relationships. Various TSS to 2D image conversion methods for NILM were tested, including Gramin Angular Summation Field (GASF), Gramin Angular Difference Field (GADF), Recurrent Plot (RP), and Markov Transition Field (MTF), with GADF outperforming other methods. In addition, the study introduces a simple yet powerful 2D input mechanism for time series data, specifically energy consumption data. This mechanism will be integrated into a CNN-based energy disaggregation model for the first time in the NILM domain, with the aim of improving overall performance. While the proposed method excels over 1D input-based models in training, it is observed that the novel 2D input method requires augmentation in training data volume, data mixing, NN depth, and hyperparameter tuning to achieve superior generalization capabilities. Furthermore, aggregate energy signal-based Voltage-Current (V-I) trajectory plots were investigated for fully non-intrusive appliance classification, demonstrating high accuracy. v The study proposes a single NN architecture named "One-Shot." This model exhibits the capability to simultaneously disaggregate multiple appliances, offering a more efficient alternative to the intricate and computationally demanding existing NN-based NILM models that necessitate separate NNs for each appliance. The efficacy of this approach is evaluated across multiple input-output mapping configurations, with the multi-point multi-bin model proving superior. To address challenges associated with manual model re-training for new appliances and adapting to evolving consumption patterns, a self-learning module is incorporated, enhancing the performance of the OneShot model. To overcome issues related to excessive hyperparameter tuning and insufficient training data, the study presents an unsupervised model based on Blind Source Separation (BSS), utilizing Independent Component Analysis (ICA) to separate appliance energy signals from the aggregate signal. Developing more reliable disaggregation models in local environments requires a local energy dataset. For this purpose, the study creates a local energy dataset from households using a custom-designed data logger, capturing both low and high-frequency energy data at appliance, circuit, and main energy meter levels. This dataset is verified using the One-Shot model developed in this study. In summary, this study advances the field of NILM by introducing AI-based solutions, innovative approaches, and comprehensive guidelines. Ultimately, these contributions aim to foster energy conservation and enhance efficiency in residential and commercial settings globally.