Master of Science in Information Technology

Permanent URI for this collectionhttp://192.248.9.226/handle/123/85

Browse

Now showing 1 - 20 of 288

item: Thesis-Abstract
IoT empowered open sensor network for environmental air pollution monitoring system in smart cities
(2023) Manchanayaka MALSK; Sudantha BH
Designing an IoT-based air pollution monitoring system is a great initiative to address the issue of air pollution and its impact on the environment. Such a system can provide real-time data on air quality and enable timely actions to be taken when pollution levels exceed acceptable limits. Here's an overview of how the system has been designed: Hardware Components: 1. Air Quality Sensors: Select appropriate sensors capable of measuring key pollutants such as particulate matter (PM2.5, PM10), nitrogen dioxide (NO2), ozone (O3), VOC (Volatile Organic Compounds) and humidity, and temperature etc. These sensors should be able to provide accurate readings. 2. Microcontroller: The Arduino Mega 2560 microcontroller board has been selected considering the open platform to interface with the sensors, process the data, and control the system's operations. 3. Communication Module: A communication module added to have GSM connectivity to transmit the collected data to a central server or a cloud platform. 4. Power Supply: As the reliable power source for the system, renewable power sources such as a 50 W solar panel is used together with a rechargeable lithium iron phosphate (LiFePO4) battery, to make it independent of the electrical grid. Software Components: 1. Sensor Data Collection: Developed firmware and software to retrieve data from the air quality sensors connected to the microcontroller. The data is sampled at regular intervals. 2. Data Processing: Implemented algorithms to process the collected data, perform necessary calibration, and convert it into meaningful air quality parameters. These parameters can include Air Quality Index (AQI) or pollutant concentrations. 3. Data Transmission: Established a secure connection between the microcontroller and a central server or cloud platform. Protocols like MQTT or HTTP were used to transmit the processed data over cellular networks. 4. Data Storage and Analysis: Stored the received data in a database for historical analysis and trend monitoring. Data analytics must be done to identify patterns, pollution sources, and correlations with other environmental factors. v 5. Alerting System: Set threshold levels for different pollutants, and when the measured values exceed these thresholds, trigger alert signals. Alerts can be sent via SMS, email, mobile app notifications, or other appropriate communication channels. 6. User Interface: Develop a user-friendly web-based or mobile application to visualize the real-time and historical air quality data. This interface should allow users to monitor air quality, view alerts, and access additional information or recommendations. Security and Privacy: Implement appropriate security measures to protect the system from unauthorized access. Ensure data encryption during transmission and storage. Follow best practices to safeguard user privacy. Integration and Scalability: Consider the scalability of the system to accommodate additional sensors or monitor air quality in multiple locations. Allow for easy integration with existing environmental monitoring networks or systems. Deployment and Maintenance: Deploy the IoT-based air pollution monitoring system in strategic locations, such as urban areas, industrial zones, or near sensitive ecological areas. Regularly calibrate and maintain the sensors to ensure accurate measurements. Update the software and firmware as needed to address any issues or improvements. By designing and implementing an IoT-based air pollution monitoring system, you can effectively monitor air quality, raise awareness about pollution levels, and enable timely actions to mitigate the harmful effects of air pollution
item: Thesis-Abstract
A Machine learning approach to assist the prediction of loan characteristics
(2022) Perera CL; Premarathne SC
The business environment in Sri Lanka has become complex and competitive with the development of the financial sector and the spread of the Covid-19 pandemic. The number of business organizations and individuals applying for loans has increased. The practices that are being used to predict financial allocation for loans of future periods are based on previous experiences and rough estimates. The most challenging risk faced during this process is the credit risk, which is the risk of lending money to unsuitable loan applicants. Lengthy authentication procedures are being followed by financial institutes prior to approving loans. However, there is no assurance whether the chosen applicant is the right applicant or not. Also, predicting the risks of credit loans prior to becoming non-performing is essential as the outcomes are unbearable except provisions are arranged for anticipated downsides. Thus, this study focused on analyzing the historical data of loans and evaluating customer profiles based on the demographic, geographical, and behavioral data of the customers to enable the prediction of future loan amounts, evaluation of the credit risks of loans and prediction of Non-Performing Loans using Machine Learning (ML) algorithms, in order to help make appropriate choices in the future. An exploratory data analysis was first performed to provide insights on developing marketing strategies based on loan types and to identify the type of customers who can be approached. Thus, three models were devised to predict the identified loan characteristics. Model 1 was devised to predict the future loan amounts with the highest R-squared score of 0.9967 using Light Gradient Boosting Regression. Model 2 was devised to evaluate the credit risk with the highest training and test accuracy of 0.9960 and 0.7842, respectively, using Stacking Ensemble Classification. Model 3 was devised to predict the Non-Performing Loans with the highest training and test accuracy of 0.9999 and 0.9522, respectively, using Random Forest Classification. Finally, the study illustrated a remarkable approach in predicting loan characteristics which ideally suits the ever changing economy. It achieved outstanding results which could enable any financial institute in the country, in minimizing the expected risks.
item: Thesis-Abstract
Decision support system to predict business performance: study of small and medium scale enterprises during COVID - 19
(2022) Jayaweera MADBL; Karunaratne I
The COVID-19 pandemic has disrupted business activity, particularly for small and medium enterprises, some of which have been entirely lost to the economy. This unexpected crisis has required that businesses change quickly with few resources. Some enterprises respond to this situation by applying creativity and have responded to the change more successfully than others. Generally, business performance depends on the owner’s characteristics and how they operate their businesses. Therefore the aim of this research was to examine the relationship between owner characteristics and small business performance and limitations during the pandemic. This research consisted of collecting primary data, using a questionnaire, from small and medium enterprises in three grama niladari divisions, in Chilaw. Hypotheses were tested using descriptive techniques, multiple regression analyses, and decision tree algorithmhem. It was hypothesized that the owner’s s characteristics would relate to business performance according to sales growth, profit growth and number of employees growth. The results indicate that a substantial proportion of businesses have closed due to the pandemic. Business owners had used financial and non-financial strategies to tackle the crisis (obtaining loans, utilizing business and social networks, pursuing new market channels), but a substantial number simply did not adopt particular strategies. The personality characteristics, adaptability, competitiveness, autonomy, risk propensity and emotional resilience significantly affected business performance (p<0.05). The owner’s age, business type, business age, and some finacila and non-financial strategies also showed significant relationships with business performance (p<0.05). The research findings give insights into how the pandemic has taken a toll on SMEs and findings are used to build up a decision support system
item: Thesis-Abstract
Decision making dashboard for agile test automation development
(2022) Mahagedara MMSU; Wijesiriwardana C
Automatic testing enables faster iterations and reliable test outputs which leads to the delivery of a high-quality product. Test Automation in Agile development improves efficiency of development and faster execution, Reusability of test component on Repetitive nature of tasks, Higher test coverage, and helps to reduce overall testing cost to provide a high quality product. In Agile development environment, developers and quality engineer’s work together and in every sprint and releases set of features to the product decided by product owners. Quality engineers in agile team covers manual testing and test automation for the features within the sprint allocated. In most of the companies agile team tracks and display manual testing progress, bugs and defects found, test coverage, test cases for the new features developing but not the test automation progress and results. In some of the startups and even in the multinational level companies, in their product teams, quality engineers doesn’t have any proper mechanism to provide test automation results in proper way. Most of the quality engineers has to execute the test automation scripts manually on the local machine and give sprint updates and test automation progress updates on based on the console results. It’s hard to explain and show the results and test automation coverage on console for the offshore product owners and the team. As a solution this application will help project managers, business analysts and offshore product owners and the whole team to make decisions based on the test results within the sprint. This proposed application will help quality engineers to update the completed test automation results on the dashboard within the sprint so management and the agile team will be able to make decisions whenever necessary. And quality engineers does not need to execute in local machine and give updates to the management and the team as it’s already updated on proposed dashboard. This proposed application would function as a decision making dashboard using test automation results for the small scale agile projects
item: Thesis-Abstract
Analyzing reasons for unanswered questions in stack overflow
(2022) Dissanayake BAK; Wijesiriwardana C
In the digital world knowledge and learning depends on the internet, and it has many advantages to human. Stack overflow is significant site to software developers as well as all Information technology users. This research proposed analyzing reasons for unanswered questions in Stack overflow platform. An end-user of stack overflow website must be analysis in various methods, the users don’t have questions from one programming language and don’t have equal knowledge, but their answer or feedback should be correct and must help to improve knowledge and resolved the issue. In this research primary dataset mainly gathered from stack overflow question database. It considers attributes of stack overflow dataset which are stored to analyze unanswered questions based on supervised learning, classification models. In Stack Overflow more than 50% questions are not frequently use when compare with other sites, as a result of heigh filtering methods. Programmers usually update knowledge by reading and answering new problems. And share knowledge with other programmers frequently Stack overflow is a famous platform among IT users as well as programers to send question and a get answer. Most of problems immediately get an answer by other users and few questions remain without answers, and it is a problem for Stack overflow platform developers to keep these questions for long time as it take disk space only , because of platform developers remove these questions after some time. This is where this research problem starts. Find reason and help users to get answer is target of this project. Taken dataset with answered as well as unanswered questions with no upvoted or accepted answers. These unanswered are from android, localization, asp.net, JavaScript, java, xml, SharePoint, c, .net, mobil, sql, python, php, c#, html, jQuery, iOS, CSS …etc areas. At once it seems every question has an answer, when look at these counts it realized that there are lot of questions without answer. This project I want to find reason why questions posted on Stack Overflow has not answered. This reason analysis focuses 14 attributes of dataset questions. Day by day it rises questions and unanswered questions different areas of user knowledge so this will reflect developers how to get a proper answer quickly and it will make live changes in stack overflow site also. Based on dataset attribute values can not to find out proper path to analysis the unanswered questions at once. It needs to get more complicated datasets in different perspective of software development area. Research has tended to focus for several clusters based on end user question, and it is easy to get efficient answer. This analysis method has used in my research work. It will be based on structured primary dataset, taken from previous research. By using this analysis users can easily predict, which questions will have answer efficiently or not answered reason
item: Thesis-Abstract
Enhancing software quality in agile software development through customer feedback & reviews analysis
(2022) Thilakarathne PRHSV; Wijesiriwardana C
The agile software development process is one of the best software development process models which includes an iterative procedure according to agile practices. Kanban, Scrum, XP, and Hybrid frameworks provide based environments for continuing the agile process. Requirements gathering and analysis phase get the major priority within each framework according to the software feasibility. The research focused on the Scrum framework’s requirements gathering and analysis process along with related problems that the software team members faced. The research objective is to provide a better solution to overcome the requirements analysis problems by using a decision support system. Most of the software teams fail to identify the enhancements from the released software version’s feedback and review analysis. The proposed novelty system aids software team members to identify the failures and related enhancements that need to be improved at the next level. The decision support system is constructed by using three separate components and each analysis helps to identify the failures and enhancements. The decision support system accurately analysis the user requirements in each phase with high accuracy. The results of the analysis provide new insight into the software engineering research area. Research phenomenon makes a coherent interrelation between software quality assurance and quality engineering.
item: Thesis-Abstract
Credit risk analysis of small and medium - sized enterprise loans
(2022) Chandrasiri GDTD; Premaratne SC
Analyzing the credit risk is important in banking systems to ensure that debtors pay the loans regularly, on schedule. The inability of managing the credit risk may lead severe losses in financial institutes. Every financial institute must predict and manage the credit risk to avoid financial crises. Hence, finding an effective method for credit risk analysis is vital. Among various types of loans, Small and Medium-sized Enterprise (SME) loans dominate since SMEs are considered the backbone of any economy. With the higher amount of SME loans, the associated risk also gets increased. SME sector is considered as risky and costly than the large enterprises. The amount of non-performing SME loans has increased at a higher pace throughout the last few quarters in Sri Lanka. Hence, this study analyzes the credit risk of SME loans by using a data set received from a financial institute in Sri Lanka. It recognizes financial and non-financial attributes affecting the credit risk of SME loans. The data mining techniques were used to analyze the data and extract knowledge. It is an emerging technology that provides significant improvements in terms of making accurate decisions. Data mining is used widely for financial analysis since it facilitates knowledge extraction from large data sets and making effective decisions. The study will help financial institutes to predict the credit risk of potential SME borrowers and avoid inefficiencies in the lending process. It identifies the credit risk of debtors in different aspects. Ultimately, it provides valuable insights into effective decision making of an economy.
item: Thesis-Abstract
3 Dimensional visualization of code smells
(2022) Hasantha PAC; Wijesiriwardana C
Bad code smells are symptoms of design flaws in source code. Several tools and approaches have been proposed for detecting and visualizing code smells. To maintain the software quality, prioritizing the identification and removal of code smells are required. Identifying the code smells using visualization will helpful developers to understand and refactor the code. This study proposes a novel 3D metaphor to detect and visualize code smells by using a combination of the code city and island metaphor visualization techniques. Proposed model identifies and visualizes the code smell at different abstraction levels in a proper understandable aspect. This model evaluates by using several open source software projects and visualizing the detected code smells in abstraction levels such as classes, methods. The proposed model will allow for more research into code smell visualization and it will keep better focus on the needs of developers
item: Thesis-Abstract
A Computerized approach to enhance learning experience of O/L students in ICT education
(2022) Attanayake NHMC; Karunaratne I
Different types of challenging activities should be given to the student by the teacher through technology and the learner should be engaged in thinking. Learners of modern computer-based technologies have significant potential to provide meaningful learning experiences for knowledge building. The purpose of this study is to suggest and develop an effective e-learning framework for strengthening the online learning environment with an e-learning model to enhance online connectivity and learning. In this study, 100 ICT students studying in Grade 11 were divided into two groups of 50 each and the governing body was given 5 TEL learning materials based on the lessons related to the GCE (O / L) syllabus. The other group learned the relevant 5 lessons using the Zoom technology using the standard classroom learning-teaching methodology. The lessons used in the TEL material were adapted to the TPACK concept, allowing the student to face the assessments given at the beginning and end of a lesson for the final assessment of the lesson. Each lesson is done at the beginning and end of the lesson. Students in the controlled group are required to engage in self-study by following the links provided in these sequential lessons. At the end of the above process, Google Forms will provide two questionnaires for the students involved in the TEL material and the parents of 20 of those students. Before starting this study, all stakeholders will attend an online consultation and conclude the analysis of qualitative and quantitative data obtained using the Weka software.
item: Thesis-Abstract
Customer churn reasoning analysis model for telecommunication industry
(2022) Thilina KGK; Wijesiriwardana CP
Customer churn is the most impactful problem in every business and industry. Therefore, every company tries their best to satisfy and maintain existing customers. Today telecommunications companies are facing this problem frequently due to increasing demand of customers every day. It is very difficult to gather new customers and need to allocate a huge cost from company revenues to acquire new customers compared to retaining the existing customer, therefore it is more important to increase their customer retention and work for that. This research is based on churn customer information and the primary objective of this research is to predict the churn reason of a given customer who has predicted to be churn using modern data analytics techniques. It include Logistic Regression, Naive Bayes, Random Forest, Decision Tree, K-Nearest Neighbor, Support Vector Machine and Gradient Boost Classifier. Further, Hybrid Model has been considered using Voting Classifier ML model. The dataset used in this research is obtained through the Data Warehouse of one of the leading telecommunication companies.
item: Thesis-Abstract
Predicting absenteeism factors in the work place through data mining
(2022) Nishantha SP; Wijesiriwardena C
Absenteeism is an employee’s absence from work. Absences of employees can have a major effect on company strategies, finances, morale and other factors. Excessive absences may influence to decrease productivity of the company. Poorly performing employees cause significant losses to the organization, and absenteeism is considered one of the factors affecting performance. Therefore, understanding the causes of absenteeism can provide organizations with competitive advantage tools and open up research areas for computers and human resources fields. The purpose of this paper is to use computerized technology to discover the causes of employee absence. This study analyzes data from the absentee database and finds several factors that have a good correlation with absentees. In addition, two data mining techniques clustering and association rule mining are used to discover factors which cause in absenteeism with high accuracy. This research paper is to create association model to predict whether find the relationship of absenteeism of employee.
item: Thesis-Abstract
Towards enhancing graduate employability in information technology industry in Sri Lanka: an ontological approach
(2022) Harshani WAR; Ahangamage S
Skilful graduates are an asset to a country, as the primary supplier for the workforce of the industries. Having an abreast set of industry-required skills implies employability regardless of the nature of the job or the field of employment. In a Sri Lankan context, the demand for graduates has exponentially increased over the past few years, specifically in the field of Software Engineering. The requirement for entry-level jobs depends on the skills possessed by employees. Thereby, the university curricula play a substantial role in the skills of the fresh graduates’ skills, accentuating the need to design the university curricula to match the industry requirement. It is identified that the key stakeholders contributing towards this is the employees (who are graduates with a degree related to software engineering), the universities offering degrees in software engineering discipline and the statutory bodies who validate, standardize, and moderate such programs and the software engineering businesses in Sri Lanka. The study will focus on different attributes and relationships of the above stakeholders and detect the gaps in the employability skills provided by the university curricula and the anticipation of the employers through an ontological approach.
item: Thesis-Abstract
A System to suggest meaningful domain names in the Domains.lk search bar
(2022) De Silva RADW; Wijesiriwardana C
For any given domain registry the domain searching feature is an integral part of the domain registration process. The Initial user interaction with the domain registry would start with an availability checkup using this domain searching feature. If that is not providing a satisfactory level of clarity about the availability of the domain or the other options a customer can select, the domain searching feature is a failure itself. Most global domain registrars have sophisticated domain searching features while CCTLD (Country Coded Top Level Domain) registries find it difficult to create ideal domain searching features as they are facing the native language problem. Since the domains are typed in using English characters so a transliteration will also be required to search the domain. This ambiguity is blocking the development of an ideal algorithm to give closest suggestions for domains is native language. In this project we are trying to find a solution for that while implementing it with the domain search. Furthermore effective utilisation of customer searches is lacking with these CCTLD registry systems. Global giants in the domain registry business use this information effectively to change prices dynamically and earn lots of profits by selling premier domains to its customers. Since CCTLD is lacking such a feature in their domain search they had to sell all the domains under basic categorization and they potentially lose much profits by selling premier domains under wrong categories. Storing the domain searches customers have done, analysing them and then generating an algorithm to categorise them would be the best way to go ahead with developing such a feature in the domain search bar. All these functionalities would be associated with developing a sophisticated domain searching feature for a CCTLD.
item: Thesis-Abstract
Mining Social sentiments for demand analysis in footwear industry
(2022) Athurupane AMANPWMRPDB; Premaratne S
The public tends to express their thoughts about particular goods and or services through popular social media networks such as Twitter and Facebook, while the firms also use social media to communicate with their consumers. As a result, this beneficial information can be used to make marketing and business decisions. However, due to the vast, noisy, and dynamic nature of these information, capturing the true public opinion has become a key challenge. Sentiment analysis is one of the methods that is employed to extract positive or negative attitudes from this social media information and thus, it has drawn the attention of the scholars during last decade. Different scholars have used a variety of techniques and methods to capture accurate results. Data mining is one of the recent approaches adopted by them to obtain better results from sentiment data. Moreover, some scholars have extended their studies to gain more insights from different topics. Such studies conducted to predict future results, analyze trends, detect anomalies etc. It can be beneficial for massive industries like footwear to understand their market through these approaches and streamline their product and service catalogs to meet the needs of their customers. This research aims to analyze previous studies conducted on this area, identify their contribution, challenges and limitations, and build a new comprehensive demand prediction model for footwear industry using data mining techniques.
item: Thesis-Abstract
Boat recognition and automated harbor management system
(2022) Weerasekara WDLS; Premaratne SC
Fisheries industry is a vital sector of Sri Lanka’s economy since it is an island surrounded by a vast ocean. Over thousands of fishing vessels are departing to the ocean within a day from harbors all around the island. All the departing and arriving fishing vessels should have gone though an ample security check by the harbor authorities one by one. But with the COVID 19 pandemic situation and the social distancing procedure, harbor authorities are facing difficulties to detect and recognize fishing vessels by getting on the boats as before the pandemic situation. Also, currently harbors are using a manual, paper-based system for recording the information on boat departures and arrivals. This leads to the inefficiency of harbor management process, delays in rescue missions and failures of security missions. To solve these problems, this paper introduces a Boat Recognition and Automated Harbor Management System (BRAHMS) which is based on YOLO (You Only Look Once) v5 algorithm. A webbased solution is provided to manage fishing boat tracking information as one deliverable of the project. Also, YOLO based desktop application to recognize boats through the registered number is given as another outcome. Final deliverable is a backend reporting solution to send boat tracking information according to daily, weekly, monthly or yearly preschedule intervals. In this system, I have implemented a novel deskewing method for the slanted license plate recognition process. The deskewing process is aimed for three main approaches as auto deskewing, manual deskewing and a hybrid deskewing which uses both auto and manual processes together
item: Thesis-Abstract
Microcontroller - based consumer level weather prediction system
(2022) Kavinda MGA; Sudantha BH
Weather conditions are important to people in planning their activities. Since weather conditions changed unexpectedly, monitoring weather conditions are very important these days. Weather stations are built to fulfil that purpose. But, if your area is not covered by that kind of weather station, you will not get accurate weather forecasts. Therefore, it can be caused to facing unexpected weather conditions. A system proposed in this paper will collect weather-related data and a live preview of those data can be seen on a display. All the outputs will display in a Nextion Display, which provides a smooth and fast response due to its internal microcontroller. Then those captured data will be logged in to a remote server. Also, this outdoor unit is powered by solar energy as a sustainable energy source. From this unit, temperature, humidity, atmospheric pressure, precipitation, wind speed and direction will be measured. Upcoming weather can be determined using the changes in atmospheric pressure. Based on those data, by using the zambretti algorithm, the upcoming weather condition will be predicted. This weather prediction system will be developed with microcontrollers which come affordable yet more powerful to handle this kind of system. Therefore, weather data will be captured by using Arduino Mega 2560 microcontroller board. Then it will be transmitted wirelessly to the indoor unit that handles the processing of those data, displays predicted data and the data logging to an Adafruit IO Internet of Things (IoT) cloud service. Those tasks will be handled by the ESP32 microcontroller board.
item: Thesis-Abstract
Automatic testing of smart speaker apps
(2022) Sandaruwani JLAIA; Mahadewa K
With the emergence of the Internet of Things (IoT), Smart Speakers open up a new world where we can talk to a machine for getting help in our day-to-day lives. The Smart Speaker Apps (SSA)s provide a user-friendly vocal experience to the customers by allowing them to dictate commands to the speaker through voice commands. Amazon Alexa is one of the most prevalent smart speakers which allows third-party developers to write SSAs called Skills. Due to the prevalence of Alexa, it has become vulnerable to security and privacy threats by malicious skill developers. In particular, Alexa skills could be overprivileged such that they collect more data than necessary or specified by the privacy policy in the skills description. In this research, we systematically explore skills to test whether the behaviors of the skills adhere to the privacy policy provided in the skill description. We extracted the utterances related to privacy-sensitive behavior of the skills through Natural Language Processing (NLP) techniques. Second, we implemented a dynamic testing tool Test case Generator & Invocator based on the fuzzing technique to automatically manipulate the inputs to the skills and observe the output to identify the skills which accept the privacy-sensitive information. During the study, we discovered that 21% of the tested skills accept privacy-sensitive data. We have simply focused on the real or actual behavior of the skills during the research. The claimed behavior of the skills is covered by our study, which will be the focus of further work.
item: Thesis-Abstract
Identifying harmful comments for Tamil language on social media
(2022) SivalIngam D; Premaratne SC
The era of social media, such as YouTube, Facebook, and Twitter adding comments to posts are being fun in the daily life of people. But this is also used to spread hate speech and organize hate based activities increasingly nowadays. Harmful and offensive text identification on social media platforms is being a trending research area over the last few years. In a country like Sri Lanka with multiple native languages, people like to comment on social media mostly in their native language. Tamil is one of the Languages commonly used and spoken in the North and East part of Sri Lanka. In recent years people like to comment not only in their native language they also comment in more than one language. In Sri Lanka, people use Singlish (Sinhala + English ) or Tanglish (Tamil + English). Because of the rapid growth of hateful content on social media, there is an immediate need for an efficient and effective method to identify harmful content. A huge number of researches have been done and are being done for automated harmful content detection online. The complication of the Natural Language constructs builds this task very challenging. A maximum of the research are done in the English Language. This research work aims to classify the code-mixed Tamil comments on social media by categorizing them as harmful and non-harmful by using machine learning models.
item: Thesis-Abstract
Handwritten sinhala character recognition using deep learning
(2022) Karunarathne ML; Wijesiriwardana C P
Sinhala language is the national language in Sri Lanka. Sinhala alphabet includes 60 characters and is slightly complex compared to other languages like English. Around 25-30 researches have been done since 1990 regarding Sinhala handwritten character recognition. Handwritten Sinhala character recognition remains mostly unsolved in pattern recognition, due to many perplexing characters and excessive curves in Sinhala handwriting. The existing recognizers are also unable to provide acceptable performance for practical applications. This research aims to enhance the performance of handwritten Sinhala character recognition by using a new approach focused on deep neural networks, which have recently given excellent performance in many applications. This research implements Convolutional Neural Networks (CNNs) and Gabor initialized Convolutional Neural Network (GCNN). In addition to that, it investigates the performance of the proposed network architectures when introducing the dropout. To apply Gabor initialized CNN, the effect of the parameters of the Gabor filter over the Sinhala character image dataset is also examined. Considering the effect of the parameter on the GCNN architecture, parameter values for the proposed GCNN architecture are determined. The training accuracy of the first CNN method is 96.33 % and the testing accuracy is 90.14%. According to the literature, this is the highest accuracy obtained for 60 Sinhala characters compared with primitive methods. This accuracy is obtained with the 0.5 dropout effect. The Gabor initialized CNN architecture provides 95.15% training and 80% testing accuracy. Even though the training accuracy is approximately 1% less than the training accuracy of the first CNN architecture, it converges to the results rapidly. So, it saves time and computational cost. Considering the results of implemented CNN architectures and Gabor initialized CNN architecture, the best-performing architecture is selected for the Sinhala handwritten character recognition process
item: Thesis-Abstract
The Guardian artificial intelligence system for roadside assistance and vehicle maintenance
(2022) Jayasuriya WGIF; Premaratne SC
Artificial intelligence for roadside assistance and vehicle maintenance system develops for users with less mechanical knowledge. The main scope of this system is to provide technical or non-technical information without delay when requested by system users. The system enables users to connect to the system using the shorts message system (SMS) feature. The system consists of 4 subsystems: the short message gateway, NLP (Natural Language Processing), Chat-bot and Database Handler. When a user sends a question to the system, it sends directly to the natural language processing module to understand what the user is requesting. If the NLP is unable to handle the problem, the system sends those requests to the Chat-bot, which can communicate with the users to handle ambiguous issues and provide solutions to the users’ problems. Existing systems for road assistance are more difficult in practice. They are more expensive to use, which means it costs more for the user to continue the service after the warranty period and they provide the service only to registered customers. Although service agents are almost always busy and need a high cell but a phone or other device to use those services. Hence, the shortcomings of existing systems lead to the development of an artificial intelligence system for roadside assistance. Furthermore, this research will help people doing research in the field of roadside assistance and also people who interest to perform their maintenance without the assistance of a third party.