Browsing by Author "Rathnayaka, R.M.K.T."

Now showing 1 - 7 of 7

Artificial Neural Network based New Hybrid Approach for Forecasting Electricity Demands in Sri Lanka.
(Faculty of Computing and Technology, University of Kelaniya, Sri Lanka., 2017) Rathnayaka, R.M.K.T.; Seneviratna, D.M.K.N.
The electricity generation and forecasting are playing a significant role to enhance national economic growth. It has a direct impact on both individual’s standards of living and industrial enhancements; especially, it is a prerequisite to enhance industrialization, farming and residential requirements. As a result, most of the countries are allocating a considerable amount for power generation and forecasting from nation’s annual budget. The main objective of this study is to focus on analyzing the electricity demands in Sri Lanka using a new proposed combined hybrid approach based on Artificial Neural Network. The methodology of the study is carried as follows. In the first phase, electricity demand of Sri Lanka is forecasting based on the autoregressive integrated moving average (ARIMA) and Artificial Neural Network (ANN) approaches separately. In the next stage, the new proposed combined approach of ANN and ARIMA (ANN-ARIMA) is applied. According to the Akaike Information Criterion, Schwarz Information Criterion and Hannan Quinn Criterion results, ARIMA(0,1,1) (R-squared : 45%, Durbin-Watson stat: 2.32) and ARIMA (1, 1, 1) (R-squared : 55%, Durbin-Watson stat: 2.03) are best models for forecasting electricity production and electricity consumption under the linear framework respectively. As a next step, proposed ANN-ARIMA hybrid methodology is applied to forecast non-linear composite based on MATLAB training algorithms. Furthermore, the model selection results concluded that, Backpropagation Neural Network (BPNN) (1-4-1) with 0.06 learning rates and BPNN (1-2-1) with 0.04 learning rates are the best one-step-ahead forecasting for electricity production and electricity consumption respectively. According to the empirical results, the electricity production and consumption curves went parallel trend up to 1995. However, after 1995 consumption rate has been increasing rapidly with respect to the production rate. When this is the case until 2020, it will create distortions in the Sri Lankan future. So this study is a good sign for the government and energy sources must be introduced and implemented for national power grid early as possible.
CNN based deep learning model for tomato crop disease detection
(Faculty of Science, University of Kelaniya, Sri Lanka, 2020) Gunarathna, M.M.; Rathnayaka, R.M.K.T.
Tomato is one of the most commonly cultivated solanaceous short duration vegetable crops, grown in outdoor and indoor conditions, worldwide. However, unfortunately, many diseases affect these crops which have an impact on quality and the quantity of the produce, agricultural productivity, and causes considerable economic losses to the producers and to the contribution to the growth of the agricultural sector. Therefore, continuous monitoring of the crop is required throughout the growing stage to identify the diseases. The most traditional way of identifying diseases is naked eye observation, which is tedious and time-consuming. Today, advances in computer vision paved by deep learning have led to a situation where disease diagnosis is based on automated recognition. The main objective of this study is to develop an accurate tomato disease classification model which eliminates human error when identifying diseases. Due to a variety of similar disease and pathological problems, even experienced agronomists and plant pathologists often fail to recognize the correct disease. This computer vision system will assist agronomists in detecting a variety of tomato crop diseases. The proposed algorithm consists of four main steps; data collection, data pre-processing, CNN model creation, and evaluation of performance metrics. A leaf is a good indicator of the tomato’s health. Therefore, tomato leaf images belonging to 10 different classes with a resolution of 256x256 were collected from the Internet to build, validate, and test the model. Collected images were normalized and image augmentation techniques were applied to increase the size of the training data set in the preprocessing phase. The CNN model of the study was built from scratch using the Keras library, which runs top of the Tensorflow backend. The model comprises four convolutional blocks followed by batch normalization, max pooling, and dropout layers. Two dense and flatten layers were also included at the end. A time-based learning rate scheduler was used with an initial learning rate of 0.001, momentum of 0.5, an epoch of 15 and a batch size of 27. The final model was able to achieve a training accuracy of 94% and a testing accuracy of 92%. This proposed system would encourage tomato cultivators to detect diseases at an early stage and start treatments without relying on experts. In the future, we hope to build an ensemble approach to classify plant diseases with real-time images towards the development of a decision support system.
Computer aided segmentation approach for Melanoma skin cancer detection
(International Research Conference on Smart Computing and Systems Engineering - SCSE 2018, 2018) Sivathmeega, S.; Kariapper, R.K.A.R.; Rathnayaka, R.M.K.T.
Skin cancer is the most common type of cancer in the world and nowadays, this incidence is increasing rapidly. In recent years, there has been a fairly rapid increment in melanoma skin cancer patients. Melanoma, this the deadliest form of skin cancer, must be diagnosed earlier as soon as possible for effective treatment. To diagnose melanoma earlier, skin lesion should be segmented accurately. However, the segmentation of the melanoma skin cancer lesion in traditional approach is a challenging task due to the number of false positives is large and time consuming in prediction. Hence, the development of automated computer vision system becoming as an essential tool to segment the skin lesion from given photograph of patient’s cancer affected area and to overcome those difficulties, which were found in the earlier methods. This work was done through image processing techniques. Some of these techniques are widely used in similar applications, as is the case of the canny edge detection for finding the lesion boundary. Other techniques are watershed segmentation for segmenting the lesion from skin, multilevel thresholding for merging the lesion, and active contour for increasing the accuracy. Though the personnel in the medical field had introduced new methodologies to improve the accuracy by addressing the challenges and mainly focusing on the accuracy, the approach in this study achieved 97.54% sensitivity, 97.69% specificity, and 97.56% accuracy.
Database Management System Deployment on Docker Containerization for Distributed Systems
(Faculty of Computing and Technology (FCT), University of Kelaniya, Sri Lanka, 2021) Kithulwatta, W.M.C.J.T.; Jayasena, K.P.N.; Kumara, B.T.G.S.; Rathnayaka, R.M.K.T.
Containerization is a novel technology that brings an alternative for virtualization. Due to the most infrastructure-based features, most computer system administration engineers use Docker as the infrastructure level platform. On the Docker containers, any such kind of software service can be deployed. This study aims to evaluate Docker container based relational database management system container behavior. Currently, most scholarly research articles are existing for the database engine performance evaluation under different metrics and measurements of the database management systems. Therefore, without repeating them: this study evaluated the data storage mechanisms, security approaches, container resource usages and container features on the launching mechanism. According to the observed features and factors on the containerized database management systems, containerized database management systems are presenting more value-added features. Hence containerized database management system Docker containers can be recommended for the distributed computer systems for getting the benefit of effectiveness and efficiency.
MLP Model Approach for Driver Fault Identification
(4th International Conference on Advances in Computing and Technology (ICACT ‒ 2019), Faculty of Computing and Technology, University of Kelaniya, Sri Lanka, 2019) Ariyathilake, S.N.; Rathnayaka, R.M.K.T.
The issue of the traffic accident has gain attention of the globe which has been a major challenge for the sustainable development of transportation and traffic. Crashes are events which occurred by involving different components: Driver, road, environment. Driver identification is directly connected to taking advanced actions on the road accident. Prevention of the road accident is the primary concern and necessary legal actions must be taken for the responsible party of the accident. In order to accurately predict the driver fault regarding an accident, this study aims to identify whether the driver is fault for the accident or not, by using a Multilayer Perceptron (MLP) model. The proposed model accurately predicts the driver fault while ensuring the accuracy of the decision. Proposed Multilayer perceptron model has achieved an accuracy of 97.77% with the accident data. To compare the results of the model, Decision Tree, Linear classifier and DNN classifier has used. Comparative results revealed that the most accurate model as the Multilayer perceptron approach. Necessary sensitivity analysis regarding the MLP was performed to find the best MLP model. Results revealed that by using 500 epochs with RMSprop accuracy was increased. T – Test was performed with 0.05 accuracy level for the selected methods and MLP method outperformed the other techniques. The research will provide the information needed to guide the relevant decision-makers in adopting suitable measures to prevent and to reduce the accident rate.
A New Financial Time Series Approach for Volatility Forecasting
(Department of Statistics & Computer Science, University of Kelaniya, Sri Lanka, 2016) Rathnayaka, R.M.K.T.; Seneiratna, D.M.K.N.; Arumawadu, H.I.
The investment in capital market is easiest, fastest and securable way for building healthy financial foundation today. Because of the economic outlooks causing directly on these market fluctuations, the making decisions in the equity market has been regarding as one of the biggest challenges in the modern economy. The main purpose of this study is to take an attempt to understand the behavioral patterns and seek to develop a new hybrid forecasting approach under the volatility. The results are successfully implemented on Colombo stock exchange (CSE), Sri Lanka over the three year period from January 2013 to December 2015. The empirical results indicated that the new proposed hybrid approach is more suitable for forecasting price indices than traditional time series forecasting methodologies under the high volatility.
An Unsupervised Machine Learning Approach for Churn Prediction
(4th International Conference on Advances in Computing and Technology (ICACT ‒ 2019), Faculty of Computing and Technology, University of Kelaniya, Sri Lanka, 2019) Prasanth, S.; Rathnayaka, R.M.K.T.; Arumawadu, H.
Customer churn is one of the critical issues faced by the firms nowadays. Telecommunication industry is no exception to this rule. In this industry, keeping the existing subscriber (customer) is more valuable than acquiring a new subscriber (attracting new customers costs approximately 5 times higher than retaining the existing customers). Therefore, predicting the attrition behavior of customers in advance is a significant task. This behavior has triggered most of the researchers to focus on developing the churn prediction model in several industries. Anyhow, in most of the time supervised machine learning techniques have been incorporated in this regard. But in here, an unsupervised machine learning approach has been proposed. A local telecommunication company can be approached for the purpose of conducting this research. Around 10,000 postpaid subscriber details with 20 attributes have been obtained and analyzed during this research. Further, Principal Component Analysis (PCA) and Kmeans clustering algorithm have been utilized with the intention of reducing the dimensionality between features and to find the churners and non-churners respectively. The results obtained from the PCA have revealed that, 16 principal components which represent all the 20 features are considered as most important aspects to cover the entire data. Moreover, totally 6 clusters have been generated and some particular features that tend to show high contributions were identified during the principal component analysis have been analyzed towards each cluster. The proposed approach has finally revealed that out of the 6 clusters three (3) representing 4888 are churners and the other three (3) representing 5112 are non-churners. It could be ensured that, this approach would assist the future researchers to have a promising start for combining the unsupervised technique with the supervised one.