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.

CC BY-NC-ND 4.0 License

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

Now showing 1 - 20 of 290

Embargo
10-Year Cardiovascular Disease (CVD) Risk Prediction of Sri Lankans: A Longitudinal Cohort Study
(2021) Solangaarachchige, M.B
Cardiovascular diseases are one of the leading causes of mortality in the world. A cornerstone of preventive cardiology is identifying individuals at risk of cardiovascular diseases (CVD) at the earliest. Clinical guideline primarily recommends risk prediction models that are based on a limited number of predictors that perform poorly across all patient groups. Predicting cardiovascular risk is crucial for making treatment decisions, especially in the primary prevention of CVDs using a total risk approach. Despite the fact that several cardiovascular risk prediction models exist, only a handful are specifically designed for Asians, and none are generated from South Asians, including Sri Lankans. Machine learning (ML) and neural networks appear to be increasingly promising in supporting decision-making and forecasting from the huge amounts of data generated by the healthcare industry. This led us to develop a CVD model using Machine Learning to predict 10-year risk of developing a CVD in Sri Lankans. We investigated whether we could adopt ML to develop a model and whether there is an improvement in including nontraditional variables for the accuracy of CVD risk estimates and how to validate the ML model with existing WHO risk charts. Using data on 2596 participants without CVD at baseline data collection of Ragama Medical Officer of Health (MOH) area in Sri Lanka, we developed a ML-based model for predicting CVD risk based on 75 available variables. However, the ratio of developing a CVD vs no CVD in 10 years was 7:93, which is extremely unbalanced. Therefore, at first, we derived a balanced dataset from the main dataset and build a ML model and it recorded an 80.56% accuracy. Secondly, to alleviate the dataset's imbalance, we adopted two techniques, which are 10-fold cross validation and stratified 10-fold cross validation (SKF) and trained six ML classification algorithms. They are Random Forest (RF), Decision Tree, AdaBoost, Gradient Boosting, K-Nearest Neighbor and 2D Neural Network. Out of these six algorithms RF model with SKF showed the highest accuracy in predicting a CVD event with an accuracy of 93.11%. Our ML model included predictors that are not usually considered in existing risk prediction models. Systolic blood pressure was the most important variable in this model. There were six non-traditional variables in the most ten important variable list and three of them were non-laboratory variables. To validate the model with existing WHO risk charts, we explored an experimental approach by developing a simple logistic regression function using the same techniques as the best selected model, with the seven traditional risk factors used in WHO risk charts and our Random Forest model indicated the highest accuracy compared to the WHO model, with a difference of 26.20 %. Our ML model improves the accuracy of CVD risk prediction in the Sri Lankan population. This approach justifies that the CVD prediction models also can be derived using ML for each subregion individually. Additionally, our research discovered novel CVD disease factors that may now be investigated in prospective studies.
Open Access
A Flask-Based System for Measuring and Analyzing Confidence in Interviewee Speech Using Speech Recognition Technology
(Sri Lanka Institute of Information Technology, 2025-12) Dangalla, H.P.
Confidence is vital in the interview process, as it is a key determinant of credibility and competence. Nevertheless, conventional approaches to evaluating confidence are highly dependent on human judgment, which brings in bias and variability. This article suggests an effective machine learning platform which uses convolutional neural networks (CNNs) and long short-term memory networks (LSTMs) to overcome these shortcomings, offering an objective and scalable method to determine confidence levels in speech during interviews. In this architecture, CNNs extract the spatial characteristics of audio spectrograms, paying attention to the key prosodic variations in pitch and tone that act as confidence indicators. Meanwhile, LSTMs learn the time-varying behavior of these features, enabling the system to identify change in speech rate and time-varying pauses. These models can jointly identify speech as confident or non-confident with 92.5 percent accuracy on labeled data. This system is more precise, recalls higher, and has a better F1 score than current methods. Although the model demonstrates potential in confidence detection, it struggles with extrapolating across accents and languages due to overfitting. But it has a lot of potential in the future as a tool. To overcome future challenges, more diverse datasets and sophisticated methods such as data augmentation and transfer learning can be implemented to enhance the adaptability of the system. Such a framework might be of immense use in practical situations when conducting job interviews, educational evaluations, and coaching in speech delivery, giving consistent, objective measures of confidence. The resultant system might help enhance fairer judgments, offer constructive criticism to applicants, and contribute to making informed choices, benefiting the science of affective computing. It also paves the way to scalable, real-time solutions that could improve human-AI interaction and enhance communication dynamics in various areas.
Open Access
A Hybrid Machine Learning and Rule-Based Approach for a Sinhala Natural Language Interface to Database
(Sri Lanka Institute of Information Technology, 2025-12) Mahdi, M. R. M.
The objective of Natural Language Interfaces to Databases (NLIDBs), is to provide users with an intuitive way to get at their data; users can ask questions of their relational databases using natural language, instead of using a formal query language like SQL. While there have been significant advancements in developing NLIDs for high resource languages such as English, support for low resource and morphologically rich languages such as Sinhala continues to be limited. Most existing Sinhala NLIDBs have employed rule-based approaches, these have limitations in terms of adapting to new conditions and scaling. In this paper we propose a hybrid approach to developing Sinhala NLIDBs that combines rule-based logic with statistical methods to address the current limitations of Sinhala NLIDBs. Our focus will be on a single student table and supporting the basic SQL operations. We will employ a combination of core linguistic preprocessing techniques (tokenization, stemming, POS-tagging) along with a grammar driven query parser that is specifically designed to accommodate the unique structure of Sinhala. We will use a manually annotated dataset of 800 Sinhala-SQL query pairs to improve our model’s ability to identify semantic elements through Named Entity Recognition (NER). Furthermore, we will employ an intent classifier to guide the SQL generation process, enabling us to correctly understand a variety of natural language queries. Our hybrid architecture seeks to achieve a balance between the precision of rule-based systems and the flexibility of statistical systems; providing both interpretability and generalizability. In addition to improving the accessibility of databases for users who speak Sinhala, this research provides a foundation for developing future multilingual and multidomain NLIDBs for low resource languages.
Open Access
A Machine Learning Approach to Identify the Key Factors Affecting Correct Stream Selection and To Predict Suitable Subject Streams for Advanced Level Students in Sri Lanka
(Sri Lanka Institute of Information Technology, 2025-12) Abeywardhana,K.G.H.
Education plays a vital role in shaping the economic growth and sustainable development of a nation. It is not only a measure of a country’s intellectual wealth but also a determining factor in its future progress. In Sri Lanka, education is provided free of charge by the government from primary school through university, ensuring equal access for all students. Within this framework, the General Certificate of Education (Ordinary Level) – G.C.E. (O/L) and the General Certificate of Education (Advanced Level) – G.C.E. (A/L) examinations represent two critical milestones in the academic journey. The G.C.E. (A/L) examination, in particular, serves as the gateway to higher education and university admission, marking a pivotal stage in shaping students’ academic and professional futures. At the end of the O/L stage, students are required to select a subject stream such as Science, Arts, Commerce, or Technology to pursue during their A/L studies. This choice has a lasting impact, as it directly determines the student’s educational direction and career opportunities. However, many students make this crucial decision based on external influences, such as parental pressure, peer comparison, or limited guidance, rather than through a clear understanding of their academic strengths, personal interests, or long-term career aspirations. Consequently, this often leads to dissatisfaction, stream switching, or even discontinuation of studies. To address this issue, it is essential to adopt a data-driven approach that considers multiple factors, including students’ O/L examination performance, inborn talents, extracurricular activities, and preferred professional fields. This research introduces a machine learning-based model the Subject Stream Prediction System—designed to recommend the most suitable A/L subject stream for students. The proposed system not only predicts the optimal subject stream but also provides additional guidance by suggesting potential career paths, relevant educational qualifications, and technical skills aligned with the student’s profile. Four supervised machine learning algorithms K-Nearest Neighbors (KNN), Decision Tree, Random Forest, and Support Vector Machine (SVM)were trained and evaluated to develop the predictive model, ensuring the highest possible accuracy and reliability.
Open Access
A Metric-Driven Framework for Evaluating Large Language Models in Software Testing: Insights from Industry Experts
(Sri Lanka Institute of Information Technology, 2025-12) Perera, V. I. T.
The integration of Large Language Models (LLMs) into software testing workflows has introduced new opportunities for automation, but also raised critical questions regarding the reliability, maintainability, and effectiveness of the generated test cases. This study addresses the lack of standardized evaluation practices by proposing and validating a comprehensive metric-driven framework: STEAM-LLM (Software Test Effectiveness Assessment Model for LLM-generated tests). Through a mixed-methods research design involving structured surveys and expert interviews, the framework identifies three core independent variables: Prompt Engineering Level, Human Intervention, and Input Specification Quality and evaluates their impact on FaultRevealing Power and Maintainability & Stability. The inclusion of Edit Effectiveness as a mediator, along with contextual moderators such as Task Complexity, Developer Experience, and LLM Class, reflects the nuanced dynamics of LLM-assisted testing. Confounding variables including Baseline Project/Test Quality and Time Spent on Understanding the Task were also accounted for to ensure valid assessments. The framework was empirically validated and supported by metric thresholds (e.g., ≥80% coverage, ≤10% smell density), offering practical benchmarks for industry adoption. The findings demonstrate that structured prompts, strategic human oversight, and high-quality inputs are essential for producing reliable and maintainable LLM-generated test cases. This research contributes a theoretically robust and practically applicable model for evaluating AI-assisted software testing, laying the foundation for future experimentation, tooling, and integration into continuous development environments.
Open Access
A Multi-Modal Deep Learning and Explainable AI Framework for Transparent Job Matching and Career Development
(Sri Lanka Institute of Information Technology, 2025-12) Warnasooriya,D. M. D. W. R
In the digital age, career and professional growth are being influenced with advanced systems which enable finding employment, reviewing applicants and skill advancement. The changing aspect of Explainable Artificial Intelligence (XAI) is essential to allow contextual job matching and reduce discrimination in AI-based job processes. However, most of the existing systems are still non-transparent and restrictive, perpetuating prejudice and weakening credibility. The study presents a new career development and recruitment platform using XAI and surpasses the traditional methods. The proposed system uses a hybrid two-stage system to combine deep learning with the Graph Neural Networks to encode candidate job relevance as well as structural dynamics of career progression, skills dependencies, and mentorship networks. New feature engineering algorithms simulate the dynamics of temporal profiles development and skill acquisition, which allow dynamic and context-sensitive candidate representations. In order to guarantee interpretability, a recruitment-specific explainability engine offers stakeholder-specific explanations such as comparative explanations between a candidate and a job, trajectory correspondence insights, and visualizations of fair trade-offs. The system is tested to execute its functions: a real-world evaluation, which is a combination of fairness statistical measures and accuracy with user-centric interpretability measures, proves the effectiveness of the system. The results highlight the potential of radically changing the current state of hybrid AI architectures and domain-specific explainability to create ethical, equitable, and adaptive solutions in the future of work.
Open Access
A Risk-Based Complexity Metric for Measuring Scope Creep’s Effect on Agile Test Automation and QA Efficiency
(Sri Lanka Institute of Information Technology, 2025-12) Karunasinghe,A. D. A. L.
Scope creep remains a persistent challenge in Agile software development, often leading to disrupted testing cycles, reduced automation efficiency, and compromised product quality. This research introduces a novel, risk-based complexity metric Scope Creep Impact Metric (SCIM) designed to quantify the impact of requirement volatility on Agile Quality Assurance (QA). The study also proposes a composite QA Efficiency Index to measure testing performance under changing scope conditions. A mixed-methods approach was adopted, combining statistical correlation analysis with expert validation and a retrospective case study. SCIM integrates eight key parameters, including frequency and complexity of scope changes, test coverage loss, defect trends, and execution delays. The QA Efficiency Index aggregates indicators such as defect count, detection rate, regression runtime, and test execution timing. The findings revealed strong correlations between SCIM and QA disruption metrics, particularly test coverage and runtime. Hypothesis testing confirmed that complexity and risk severity are more predictive of QA inefficiency than change frequency alone. Based on these insights, an Adaptive Scope-Creep Risk Framework for Quality Assurance (ASCR-QA) was developed and validated through simulation. The framework enables Agile teams to detect scope-related risks early, adapt testing strategies, and allocate resources effectively. This research contributes a structured, empirically grounded methodology for managing scope creep in Agile environments. It offers both theoretical advancement and practical tools for improving QA resilience, supporting more predictable and high-quality software delivery.
Open Access
“AccessBIM” - A Model of Environmental Characteristics for Vision Impaired Indoor Navigation and Way Finding
(Curtin University, 2018-06) Jayakody, A
The navigation of indoor and outdoor environments play a pivotal role in the daily routine of humans. Navigation systems that provide path planning and exploration services for outdoor environments are readily available while navigation within a building is still a challenge due to limited information availability and the poor quality of GPS signals, which makes it difficult to capture characteristics within the indoor environment. Consequently, the use of GPS tracking devices for real-time map generation is not feasible. Indoor navigation is particularly difficult for people with vision impairment. According to the factsheet of the World Health Organization (WHO) as of October 2017, over 253 million people are estimated to be vision impaired: 36 million to be blind, and 217 to have poor vision. Currently, most blind and vision-impaired individuals use the white cane as an assistive tool and are often accompanied by care takers or voluntary helpers. Most modern indoor environments consist of complex architectural structures with varying arrangement of physical objects. Since retrieving indoor location information has been challenging for the vision impaired, it would be helpful if spatial information of doors, walls and staircases were made available. To address the above-mentioned problem, this thesis presents an improved schema design, an Accessible Building Information Model (AccessBIM) which could be used for generating an indoor map that could instruct vision impaired individuals in navigation, by the classification of real world objects and their locations. AccessBIM is a real-time relational database, which acts as the main component of the central system implemented to manipulate crowdsourced data such as the floor plan and architectural data along with semantic information within the built environment. The AccessBIM database stores information on the indoor arrangement of objects within buildings to facilitate the exchange and interoperability of real-time information. The database is equipped with an optimization algorithm that reduces the query execution time with the support of indexing, query re-writing, schema redesigning and a memory optimization technique introduced as “BIMcache”. vi In order to create a real-time map, the AccessBIM manipulates crowdsourced data from “smart devices” or AccessBIM users. The collection and storage of crowdsourced data, database optimization, API functions and the map construction algorithms were tested using a simulated test engine. The AccessBIM framework has the potential to play an integral role in assistive technologies related to localization and mapping, thus significantly improving the independence and quality of life for people with vision impairment whilst also decreasing the cost to the community related to support workers
Open Access
Address IoT Security and Privacy Challenges
(2021-01) Wijesinghe, K.I
Data Innovation has seen quick cross-stage and crosses utilitarian improvements for example sensors, Nano-innovation, and bio-enterprises. In medical clinics, for the most part, the E healthcare framework is utilized for getting the data of a patient. Outstandingly, the living e healthcare approach has been achieved inside cabled discussion among recognized fields for example network convention and data set in hospice climate. There has been an expansion in the healthcare framework's utilization of the versatility attributes and remote correspondence and the rise in advancements has empowered shrewd apparatuses and devices with mean evaluating energy to take advantage of remote sensor hubs. In the new age of innovation and remote correspondence, the gigantic ascent in electronic devices made by advanced cells and tablets has turned into the most famous and key apparatus of everyday life. Progressions in the Internet of Things (IoT) are generally utilized for interfacing various devices like sensors, apparatuses, vehicles, and different articles. This multitude of devices might furnish with radio-frequency identification (RFID) tags, actuators, sensors, cell phones, and numerous others. By utilizing IoT this large number of devices are associated with laying out the correspondence among themselves and effectively accessing the data. The principal favor of IoT is to enlarge the profit of the Internet with controller ability, information sharing, timeless network, and more. The healthcare servers keep electronic medical records of enlisted clients and offer various types of assistance to patients, medical advisors, and casual guardians. The patient's specialist can get the information from the office through the internet and look at the patient's set of experiences, current side effects, and patient's reaction to a given treatment. When the WBAN network is arranged, the healthcare server deals with the organization, dealing with channel sharing. A Wireless Body Area Network (WBAN) encompasses small and keen systems or contraptions subsidiary to the body of the cases moreover is to be continually managed by the cell well-being plan across a linkless discussion gear which can be Bluetooth, Zigbee, or RFID. The WBAN bargains the steady data and managing and genuine period diagrams and reactions to the business, human case, or the medical care specialists allocated for that case. Later counts seized are used for gauging clarification. The weighted counts are used to evaluate such all accommodating of disease will happen. The information is noted for the drawn-out period. Kevin Ashton first introduced the Internet of Things (IoT) in 1999. He connected numerous sensors to actual objects and relayed the collected data to the internet. The IoT mechanical talent is presently used in specific fields, such as computerized oilfield, home, and construction mechanization, smart network, improved clinical cure-wise haulage, and so on. RFIDs allow radio frequency labels to detect real counters. An RFID sensor also transmits data to the user and allows for the identification, tracking, and grouping of items. IoT science can yield colossal information about individuals, time, things, and space. Indeed, even joining the current Web science and IoT characterize an extreme use and immense amount of area set on base charge sensors and wireless correspondence. Internet convention v6 and Cloud help the advancement of a blend of web and IoT. It is enriching additional possibilities for information collecting, data treatment, organization, and different novel administrations. IPv6 is utilized to perceive an item that interfaces with IoT by an interesting addressing plan. In a country area, the majority of the people groups don't get suitable ways to deal with well-being observing and centers. Thus, it is important to plan the successful well-being observing framework. A minuscule wireless device is goal-bound with IoT can shape a possible method for directing patients remotely as opposed to dating the genuine center. The surprising little transducers are relocating into the human to total the subtleties through which the framework gets human wellness information security and for examination for treatment. The gathered information is then shipped off to remote stations through dissimilar correspondence advances (like a 3G/4G empowered base station or Wi-Fi network with the Internet. From the information that came from the internet, the medical professionals can hold onto the end and thus outfit benefits midway. The main advantage of this electronic healthcare is that it enhances the five-star exhibiting presence and offers heavenly leisure to patients and healthcare donors. The patient's privacy isn't considered in this computerized healthcare system, even though it is crucial in the patient's case, and this is its worst flaw. RFID technology is employed to overcome this problem. With its simplicity and adaptability, it handles patient reports. Similar to this, RFID's main advantage is that it defends against a variety of threats, which reduces the amount of noise in signal transmission [1][5]. A large portion of the plan is the different security systems with privacy conventions and minimal expense for improvement of materialness. Along these lines, it is important to plan useful super lightweight cryptographic conventions for a costless RFID framework. The IoT is the best answer for this reason lately. Hence, in this paper, the compelling healthcare checking framework is planned by utilizing the IoT and RFID labels. The trial brings about this paper shows the hearty result against the various attacks. In this framework to get the careful valuation return, administering and looking at the wellness state of the patient and to build the force of IoT, the blend of microcontroller with sensors is present. The various sensors are utilized to quantify the various boundaries [6]. These sensors are an ECG sensor, Pulse sensor, Temperature sensor, Movement sensor, EEG sensor, and Blood Glucose sensor. To get the productive result the blend of brilliant sensors with microcontroller parts is thought about because it enjoys loads of benefits like mean power workout, consolidated exactitude-simple abilities, and well-disposed UI. On the planet, most clinic clients utilize the PDAs and late well-being in no way, shape, or form shape administration of advanced cell sensors to administer patients' conditions. Accordingly, in this paper, there is the advantage of living advanced cell sensor devices to manage e-wellbeing. The proposed paper presents the stage for substantial sensors, which are connected straight with the patient's advanced cell to get in an arrangement at run time. This data is handled and put away in the distributed storage. The put-away data may likewise be gotten to through professionals and medical staff, later on, to notice and show victims' prosperity. Association of this paper is in the accompanying manner segment II audits the writing review of the proposed framework. In area III the Presentation of IoT and RFID are presented. Segment IV shows the advancement of the framework, and the different proposed techniques utilized in this paper are presented in this part. In segment V the exploratory execution results are presented. And at last, segment VI finishes this paper.
Embargo
Adoption of Digital Government Technologies by the Government Organizations in Sri Lanka
(2021) Mudalige, Mudalige
Moving towards Digital Government is vital for any country as it enhance the citizens’ trust towards the government through transparency in activities by applying good governance principles. As a developing nation, Sri Lanka has already moved forward from E-Government to Digital Government geared up with strengthening Digital Government policies. Government organizations are the focal point of Digital Government. For the nation to be successful in Digital Government the adoption of the Digital Government Technologies by the government organizations is vital. This study will present an assessment model to figure out the adoption level of Sri Lankan government organizations towards the Digital Government Technologies. In developing this model few existing Digital Governments in the world are studied and analyzed in depth and many inputs will be taken from these existing country specific new dimensions in assessing the adoption of the Digital Government Technologies in the Government Organizations.
Embargo
Adoption of e-business for the success of the fashion industry in Sri Lanka
(2020) Vijayanthi, H.G.S.
The purpose of this study is to investigate the adoption of e-business for the success of the fashion industry in Sri Lanka. It sets the light on why the public is reluctant in the adoption of e-business for the fashion industry even during the pandemic. The research was conducted from a deductive approach. Hypotheses were designed based on a rigorous review of the literature. The study adopted a deductive approach having the stratified convenience sample from several areas in Sri Lanka. The data obtained from the survey was analyzed through regression analysis to identify the impact of the variable on the adoption of e-business for the fashion industry. The study concludes that customer attitudes, customer abilities, ICT secure and awareness were the most influencing factors of adoption of e-business for the fashion industry according to the customer perspective. All previous studies referred to in this study have been carried out in the pre-COVID19 environment. The present study is also significant since the study also considers the effects of the pandemic. The study argues that the contradictory nature observed in perceived ease of use and perceived risk, security, and privacy concerns which were well established in previous studies is caused by the pandemic. The study was also significant since computer and internet literacy has not been tested in the Sri Lankan context in the near past and leads for recommendations that would enhance e-business to the successful fashion industry.
Embargo
The adoption of the Sri Lankan corporate sector towards digital banking during the COVID - 19 period
(2020) Deshappriya, S.K.
The pandemic of COVID- 19 can be cited as a serious challenge faced by the countries. Social mobility and physical restrictions imposed to control the spread of the COVID-19 pandemic have made it difficult for banking to function as usual. This applies to private banking as well as the corporate sector. Maintaining social distance is very important in terms of Health regulations. In such a background, it is very unfavourable to physically come to the bank branch and make transactions. Therefore, the need for working digital banking has always been acute. Banks for business and Business for banks are two indispensable sectors. Public and private banks are making strenuous efforts during this period to keep the banking business afloat. Many people think that digital banking is the best solution for this. The main objective of this study is to identify the factors influencing the Sri Lankan corporate sector to move towards digital banking. In the Sri Lankan context, there is also a research gap due to minimal publications on the subject. The quantitative research methodology has been used for this study. An online questionnaire for data collection was primarily distributed to the corporate sector. This research is based on five hypotheses. This research has been carried out by selecting a sample covering various areas of the corporate sector in the Colombo District.
Open Access
Advanced Collaborative BI Chatbot for Enhanced Enterprise Decision-Making
(Sri Lanka Institute of Information Technology, 2025-12) Shanika,W. D.
This research outlines the creation of an innovative collaborative Business Intelligence (BI) chatbot aimed at improving enterprise decision-making by utilizing context awareness, multimodal data integration, predictive analytics, and real-time collaboration. The system merges structured data from SQL databases and CSV files with unstructured resources like text files, PDFs, and images. A multimodal data integration component utilizes FAISS-based vector embeddings to enable semantic retrieval from unstructured materials while guaranteeing smooth access to structured data repositories. The predictive analytics feature goes beyond simple regression by integrating statistical model selection to determine the most appropriate forecasting technique. It also produces interactive dashboards using Dash and generates static PDF reports to cater to various decision-making scenarios. The context-awareness module incorporates tokenization, categorization, and embedding-based retrieval, as well as the capability to create user-specific reports, providing responses that are tailored to both inquiries and analytical requirements. Real-time collaboration among teams is facilitated through connections with Slack and Telegram, in conjunction with a custom chatbot interface, allowing several users to question, share, and annotate insights together. To enable enterprise deployment, the system encompasses API generation with secure management of API keys, credit allocation, and token-based pricing, ensuring controlled access and scalability. Together, these advancements transform the chatbot into a flexible decision-support platform that consolidates various data sources, generates predictive insights, produces contextual reports, and promotes collaborative analytics in real time.
Open Access
ADVANCING RANSOMWARE DETECTION SYSTEM USING MACHINE LEARNING
(Sri Lanka Institute of Information Technology, 2025-09) De Silva, G.A.A.I.S
Ransomware attacks pose a significant and evolving threat to data security and operational integrity, necessitating advanced detection mechanisms. This project aims to develop an effective ransomware detection system leveraging machine learning techniques, specifically Recurrent Neural Networks (RNN) and auto encoders, to analyze network traffic for anomalies indicative of ransomware activity. Utilizing the UNSW-NB15 datasets, we undertook extensive data preprocessing, including handling missing values and normalizing features, to prepare the datasets for training. The model employs Long Short-Term Memory (LSTM) layers to capture temporal dependencies and patterns within the network traffic data. The training and validation processes focused on normal traffic data to establish a baseline for detecting deviations caused by ransomware. Our results demonstrate high accuracy in distinguishing between normal and ransomware-infected traffic, with a clear ability to identify potential threats in real-time. This innovative approach showcases the potential of RNN-based auto encoders in enhancing cyber security measures. The conclusion emphasizes the system’s effectiveness in providing early warnings of ransomware attacks, thereby significantly aiding in the protection of valuable data assets and maintaining operational continuity.
Open Access
AgriSense: An IoT-Integrated Crop Recommendation and Price Forecasting System Using Machine Learning
(Sri Lanka Institute of Information Technology, 2025-12) Godfrri Croos, N
Sri Lankan smallholders make planting and selling decisions in the presence of shifting monsoon patterns and volatile local markets. AgriSense is an end-to-end system that integrates a low-cost IoT field device with cloud-based machine learning to support both crop selection and price planning. The device collects plot-level measurements of soil chemistry and condition, together with ambient data and location, and streams these to a backend designed to tolerate intermittent rural connectivity. In the cloud, a supervised crop-recommendation model trained on a soil feature set produces a ranked shortlist of suitable crops with calibrated probabilities.
Embargo
An AI Bot Who Is Suggesting Words to Create Trending Social Media Posts
(2021-06) Rajapaksha, S.J.S.U.
Today, social media is the mainstay of many advertising campaigns. As an example, facebook, youtube social media are widely used by TV and radio channels to advertise their programs. Not only that but many higher education institutions and even business organizations use social media extensively to reach out to a wider audience. At the same time, more and more of these advertising agencies are emerging than ever before. Despite spending so much money on social media, it can be seen that only selected advertisements reach the masses. The main reason for this is that although many people advertise on social media, they do not have a good understanding of how to do it correctly using the correct keywords. As a solution to this, before placing such an advertisement or any post on social media, if there is a prior understanding of how the product or service should be advertised these days and what words and pictures should be used for the post, then advertising is most effectively can be done. Therefore, the purpose of this project, which author is going to carry out, is to create a website for those who want to study the trending information in the social media, using artificial intelligence technology and want to do a new publicity. In other words author has created a system to get suggested keywords for social media posts according to the relevant category what should be included in their new posts to be a trending post. The author has collected data & information to archive this task related to trending social media posts for various categories and then has done a prediction by considering the amount of reaches, likes and also the comments. In this project author has used two models with linear regression and multi linear regression based AI techniques. In methodology chapter author has clearly mentioned about them. After used newly created system it was identified that new system has an increment of user reactions than using the traditional posting methods. Therefore following the testing of the new system, user responses to posts created using keywords derived from the new system were found to be higher than the responses to posts created in the normal way, and more details are contained in testing & evaluation chapter.
Open Access
AI Driven Job Recommendation System on Education and Personal Preferences
(Sri Lanka Institute of Information Technology, 2025-12) Fathima Shukra K S
Traditional career advising approaches have many disadvantages, including a narrow focus on individual assessments, standardized testing, ignoring unique circumstances of individuals, or assessment of labor market contexts or individuals. This paper describes the design and validation of an artificial intelligence (AI)-based career recommendation system that incorporates contextually relevant information including labor market trends and an individual’s educational history, abilities, work experience, and preferences to deliver the most relevant career advice. The system uses a hybrid recommendation approach that utilizes both content-based and collaborative filtering, examining the attributes of the individual as well as information about similar users that diagnostic contexts. The recommendation system uses current labor market information that is accessed via available APIs and web scraping, providing current information relevant for the recommendations. The research uses a quantitative and experimental research design with about 1,000 participants that included students, recent graduates, and early-career workers. The system is powered by machine-learning algorithms including a neural network, decision trees, and ensemble approaches and evaluates using various understanding methods for understanding performance including precision, recall, and F1-score. The system was developed as a web application using Flask, and enables users to easily enter data, see visual recommendations, and provide feedback to improve evaluation. The final contributions of this research are the ability to deliver better approach to career decision-making using form of personalized advice, provide career development planning that aligns to market conditions, provide improved generic educational planning insights, and validate a hybrid way of filtering career information for use in career guidance.
Open Access
AI for Legal Domain Identification and Guidance in Sri Lankan Civil Law: A Comparative Study of Open-Source vs Proprietary AI Models
(Sri Lanka Institute of Information Technology, 2025-12) Athulathmudali A. M.
This study investigated the application of retrieval-augmented generation (RAG) architectures powered by large language models (LLMs) to improve access to civil-law information in Sri Lanka. It addresses a key challenge in the country’s justice system, the limited accessibility to affordable and reliable legal guidance. A RAG-based legal information assistant was designed, implemented, and evaluated using two back-end models: OpenAI’s GPT-3.5-Turbo and the open-source Mistral-7B-v0.1. Both systems were integrated into a curated Sri Lankan civil-law corpus and compared across three metrics: accuracy, latency, and cost using a set of test queries. GPT-3.5 Turbo achieved higher accuracy (92.5%) and lower average latency (4.17s) at a lower cost (USD 0.000487 per query) than Mistral-7B-v0.1 (82.5% accuracy, 15.64s average latency, USD 0.000742 average cost). Statistical tests confirmed significant differences in latency and cost. GPT-3.5-Turbo therefore exhibited superior responsiveness and efficiency for real-time, citizen-facing legal assistance, whereas Mistral-7B offers a competitive, viable, privacy-preserving alternative for institutional or offline use. The research contributes a reproducible evaluation framework for legal-domain LLMs and a localized civil-law corpus designed for retrieval-augmented systems. More broadly, it demonstrates that responsibly designed AI can enhance access to justice in low-resource contexts. The findings establish a foundation for future multilingual, ethically aligned, and jurisdiction-aware legal-AI systems in Sri Lanka.
Open Access
AI Powered Log Analysis and Threat Detection System for Windows
(Sri Lanka Institute of Information Technology, 2025-12) Sriharan, G
The increasing volume and complexity of cyber threats demand advanced, automated methods for analyzing Windows event logs. Traditional rule-based systems often fail to detect novel attacks, prompting the exploration of deep learning techniques. This research develops and evaluates an anomaly detection system by fine tuning a BERT (Bidirectional Encoder Representations from Transformers) model on the windows system security logs. The methodology involved processing the ATLASv2 dataset, a collection of 20.5 million realistic Windows Security Logs containing both benign and malicious activity. A baseline model was implemented using the Hugging Face transformers library and trained on a representative sample of 100,000 log events, accelerated by a GPU. Evaluation of this baseline model on an unseen validation set demonstrated strong performance, achieving 96.98% overall accuracy and a 94.55% precision rate. The key finding was a recall of 79.10%, indicating a weakness in detecting rare malicious events due to the natural class imbalance of the dataset. To address this, a new, perfectly balanced dataset was created using oversampling, which dramatically improved the model's F1-Score to 95.33%. Following this data-centric improvement, a comprehensive hyperparameter tuning phase was conducted, employing Grid Search, Random Search, and Bayesian Optimization. This optimization successfully identified a BEST model with a high F1-Score of 96.60%. This research successfully validates a complete framework for applying and optimizing advanced AI models for log analysis. The next phase will focus on implementing a functional prototype with a user interface and expanding the comparative analysis to include other traditional ML models to further strengthen the research findings
Open Access
AI-Driven Adaptive UI Generation: Personalizing E-Learning Interfaces Based on Cognitive Abilities of Undergraduates
(Sri Lanka Institute of Information Technology, 2025-12) Weerakoon S. D.
E-learning platforms often adopt uniform interface designs and neglect learners’ cognitive differences, leading to cognitive overload and disengagement. While content personalization is common, dynamic interface-level adaptation remains underexplored. To address this gap, this study introduces an AI-driven adaptive user interface framework to personalize the interface dynamically based on individual cognitive attributes, namely attention span, memory capacity, and cognitive load, with the consideration of layout modification, navigation structure, and information density. Three validated methods are used for cognitive profiling, namely attention via WebGazer.js, cognitive load through the N-Back test, and memory capacity via the Digit Span Test. A within-subjects experiment was conducted by using 30 undergraduates in Sri Lanka. All the participants interacted with both static and AI-driven adaptive interfaces, along with a post-interaction evaluation based on validated instruments combining NASA-TLX, SUS, and UEQ scales. Results indicated a 96.7% adaptation success rate, along with positive post-feedback evaluations (M > 4.2/5) across cognitive load, navigation efficiency, personalization, usability, and engagement. Correlation patterns indicated that cognitive profiles influenced perceived outcomes. The impact of the AI-driven adaptive user interface is evaluated using quantitative analysis along with statistical data analysis using statistical software. The proposed system is designed as a web-based platform, ensuring AI-driven personalization to enhance user engagement and learning effectiveness. Further, the research findings contribute to the field of Human-computer interaction and the domain of education by validating AI-driven adaptive UI generation.