Smart Computing and Systems Engineering - 2020 (SCSE 2020)

Permanent URI for this collectionhttp://repository.kln.ac.lk/handle/123456789/23064

Browse

Now showing 1 - 20 of 37

Affective gaming in real-time emotion detection and music emotion recognition: Implementation approach with electroencephalogram
(Department of Industrial Management, Faculty of Science, University of Kelaniya, Sri Lanka, 2020) Kalansooriya, Pradeep; Ganepola, G.A.D. ,; Thalagala, T.S.
Affective Gaming can be considered as the concept of detecting the real-time emotional state of a player during various stages in gameplay and then enhancing the user interactivity accordingly to the emotional state. Based on this conception, this paper presents the research phase of the development of an Affective Car Racing computer game. The designs were created based on the theory of “Affective Loop” in games. Affective Loop consists of Emotion Elicitation, Emotion Detection/Modelling and finally Emotion Expression by Game Engine. This paper considers the second and third subphases of this loop. Designs are done for these two phases based on technologies that are still not been utilized by many game developers when designing a game. Emotion Detection/Modelling phase is introduced with a technique of capturing Electroencephalography (EEG) signals for predicting the real-time emotion of the player while interacting with the game engine. Emotion Expression phase considers the concept of Music Emotion Recognition (MER), which is a novel concept for the Gaming Industry. The authors had trained SVM models for emotion modeling via EEG Signals that will be captured by the Emotiv Epoc 14 channel device. The authors had classified the Rock and Electronic genre of music via Multi-Label RAKEL classification (Precision score of 75%) to play music excerpts based on the effect of the gamer during gameplay.
Anomaly detection in cloud network data
(Department of Industrial Management, Faculty of Science, University of Kelaniya, Sri Lanka, 2020) Yasarathna, Tharindu Lakshan; Munasinghe, Lankeshwara
Cloud computing is one of the most rapidly expanding computing concepts in the modern IT world. Cloud computing interconnects data and applications served from multiple geographic locations. A large number of transactions and the hidden infrastructure in cloud computing systems have presented a number of challenges to the research community. Among them, maintaining the cloud network security has become a key challenge. For example, detecting anomalous data has been a key research area in cloud computing. Anomaly detection (or outlier detection) is the identification of suspicious or uncommon data that significantly differs from the majority of the data. Recently, machine learning methods have shown their effectiveness in anomaly detection. However, identifying anomalies or outliers using supervised learning methods still a challenging task due to the class imbalance and the unpredictable nature and inconsistent properties or patterns of anomaly data. One-class classifiers are one feasible solution for this issue. In this paper, we mainly focused on analyzing cloud network data for identifying anomalies using one-class classification methods namely One Class Support Vector Machine(OCSVM) and Autoencoder. Here, we used a benchmark data set, YAHOO Synthetic cloud network data set. To the best of our knowledge, this is the first study that used YAHOO data for detecting anomalies. According to our analysis, Autoencoder achieves 96.02 percent accuracy in detecting outliers and OCSVM achieves 79.05 percent accuracy. In addition, we further investigated the effectiveness of a one class classification method using another benchmarked data set, UNSW-NB15. There we obtained 99.10 percent accuracy for Autoencoder and 60.89 percent accuracy for OCSVM. The above results show the neural network-based methods perform better than the kernel-based methods in anomaly detection in cloud network data.
Aspect-based sentiment analysis on hair care product reviews
(Department of Industrial Management, Faculty of Science, University of Kelaniya, Sri Lanka, 2020) Kothalawala, Malki; Thelijjagoda, Samantha
Nowadays, with almost everything being shared online, people are more verbal about their consumer experiences with products via reviews. Reviews can be vital for manufacturers to get insights into consumer opinions and consumers in their purchase decisions. Sentiment analysis, referring to the extraction of subjective opinions on a particular subject within a text, is a field within Natural Language Processing, that can convert this unstructured information hidden within reviews into structured information expressing public opinion. In regards to a specific product group like hair care products, certain brands are rising in the market due to their positive public opinion on particular aspects. While ecommerce websites facilitate users to view the reviews, they do not display which reviews contain which type of opinion on which aspect at a glance. This research aims to introduce an automated process that focuses on determining the polarity of online consumer reviews on different aspects of hair care products by using Aspect-based Sentiment Analysis. The system consists of processes like data gathering, pre-processing, aspect extraction and polarity detection and follows a sequential approach to achieve the intended goal. Consequently, by deciphering the aspect-wise polarity of the reviews, the implemented system demonstrates an accuracy of 85% from the test data for overall aspects, enabling consumers to get an at a glance idea about the public opinion and manufacturers to identify their strong and weak points.
Blockchain-based distributed reputation model for ensuring trust in mobile adhoc networks
(Department of Industrial Management, Faculty of Science, University of Kelaniya, Sri Lanka, 2020) Peiris, P.P.C.; Rajapakse, Chathura; Jayawardena, B.
Mobile ad-hoc networks also known as MANETs have been in global use for numerous applications which are not possible with fixed network topologies. The distributed operation and dynamic topology have encouraged MANETs to be applied for establishing communication in unstable environments. MANET's dynamic topology and mobility have been very advantageous in the fields of military and disaster management. These dynamic characteristics of a MANET also create a major challenge in managing trust between the mobile nodes. Managing the trustworthiness of information that a node provides to the rest of the MANET is very crucial as misinformation spread can lead to erroneous decision making. Although previous studies have been carried out on trust management in MANETs using price-based and reputation systems, the potential of a globally distributed system has not been utilized practically. Therefore, these systems address the trust management issue within a boundary of a single MANET. Above mentioned systems should be re-evaluated when a node from another MANET joins a new MANET as the reputations of the node in the previous MANET cannot be imported to the new MANET. Lack of a possible solution for this gap may result in misinformation spreading by a malicious node before other nodes determine its reputation, which could be very dangerous in sensitive environments. Therefore, a globally distributed reputation model is a timely need in mobile ad-hoc networking. Blockchain technology is one of the most suitable technologies in present for its immutable and distributed properties to build robust systems. Blockchain is a distributed ledger, that has the ability to store feedback from mobile nodes about the accuracy of information provided by other nodes. A trust factor for each node can be calculated using these feedbacks. A mobile node can then decide whether to trust information, based on nodes’ trust factors. Adopting a development-oriented research methodology, a blockchain based reputation model prototype has been implemented and validated within the study.
Contextual assistant framework for the Sinhala language
(Department of Industrial Management, Faculty of Science, University of Kelaniya, Sri Lanka, 2020) Dasanayaka, D.D.S.S.; Warnajith, N.
Continuous customer relationship plays an important role in the success of any business milieus in today’s world. Nonetheless, it can be harder to achieve consistent engagement with the customers round the clock and therefore many businesses have paved their focus in using a variety of solutions in overcoming this scenario. Contextual assistants that can have both linear and non-linear conversations with humans implicitly plays a prominent role in such situations. In contrast to resource-rich languages, creating a contextual assistant for resource-poor languages like Sinhala has been difficult mainly due to the unavailability of a rich digital footprint and the complexity of the language. Hence, this research was conducted to propose and implement a novel and common architecture of a contextual assistant framework for the Sinhala language. Here we have used a deep learning Intent Mapping (IM) model to map the consumer response to a predefined “Intent” and a Feature Extraction Mechanism (FEM) to extract related information from the input text. A set of data types for this framework were defined and FEM was trained to identify them efficiently. The IM model gave an accuracy output of 89.67 percent. The results depicted that the implemented system performs with higher accuracy in linear conversations.
Contribution of reflection in language emergence with an under-restricted situation
(Department of Industrial Management, Faculty of Science, University of Kelaniya, Sri Lanka, 2020) Todo, Kense; Yamamura, Masayuki
Owing to language emergence, human beings have been able to understand the intentions of others, generate common concepts, and extend new concepts. Artificial intelligence researchers have not only predicted words and sentences statistically in machine learning but also created a language system by communicating with the machine itself. However, strong constraints are exhibited in current studies. Models dependent on task settings, or supervisor signals and rewards exist, thus hindering the emergence of languages like the real-world. In this study, we improved Batali and Choi et al.’s research and attempted language emergence under conditions of low constraints such as human language generation. We developed a new language emergence agent that combines a language module and a visual module and included the bias that exists in humans as a “reflection function” into the new emergence algorithm. We used the MNIST dataset for language emergence. Irrespective of the function, messages corresponding to the label of MNIST could be generated. However, through qualitative and quantitative analysis, we confirmed that the reflection function caused pattern structuring in the message. This result suggested that the reflection function performed effectively in creating a grounding language from raw images with an under-restricted situation like the human language generation.
Demystifying the concept of IoT enabled gamification in retail marketing: An exploratory study
(Department of Industrial Management, Faculty of Science, University of Kelaniya, Sri Lanka, 2020) Jayasooriya, Shalini; Alles, Tharindhie; Thelijjagoda, Samantha
The retail landscape is evolving rapidly as firms embrace innovative technologies in an attempt to stay ahead of the aggressive competition prevalent within the industry. Gamification is one such innovative technology that has been gaining popularity in recent times. This study aims to explore the application of Gamification in the context of Retail Marketing in Sri Lanka and ultimately propose a concept for a Gamified application that can be used by customers of moderntrade retailers. The study took an exploratory qualitative approach where intensive surveys of literature and in-depth interviews with a judgmental purposive sample of seven marketing professionals in the modern-trade retail industry were conducted to determine the current play of technology in retail marketing as well as the drivers & challenges of Gamification adoption. Further, in-depth interviews with the customers of such organizations were conducted in gathering user preferences and design recommendations for a Gamified app. Thematic analysis was carried out in deriving insights. Findings show that the retail firms currently employ several technologies in line with those discussed in existing literature such as loyalty card systems, digital signage, VR technologies, online Gamification amidst others in carrying out their marketing efforts. Gamification is predominantly applied in the online context as opposed to the offline (in-store) context. Furthermore, the key drivers that propel firms to implement novel technology like Gamification are to generate customer insights, enhance customer experience and achieve marketing related KPI targets. Conversely, inadequate technology infrastructure, justifying the focus on a niche crowd of techoriented customers and slow ROI pose as challenges in the process of Gamification adoption. Three main themes emerged upon exploring user preferences and design recommendations for a Gamified app and are identified as information at the touch of a fingerprint, automation & integration and use of game mechanics. Ultimately by incorporating these insights gathered, a concept for a Gamified app was proposed.
Dengue mosquito larvae identification using digital images
(Department of Industrial Management, Faculty of Science, University of Kelaniya, Sri Lanka, 2020) De Silva, W.D.M.; Jayalal, S.
Dengue is one of the highest spreading mosquitoborne diseases in tropical and subtropical regions all over the world. This disease is mainly spread by the mosquito vector called ‘Aedes’. In Sri Lanka, the number of infected patients reported is increasing, and it has become a public health problem. Health Inspectors are using different methods to reduce the spread of this viral disease and one of the main methods used is the fumigation by identifying the Aedes Larvae breeding locations. Currently, this identification is done manually by the specialized health inspectors and it is totally observer-biased and consumes a considerable amount of time, which could lead to false decision making and inefficient identification. The purpose of this research is to build an automated computational model to identify Aedes Larvae in real-time with more accuracy and convenience. Even though there are good results in previous researches done in Convolutional Neural Networks (CNN) on Aedes Larvae identification, the method of capturing Larvae Images is a bit complicated since they have used a Microscope lens of amplification capacity 60-100 times to get the magnified images. In this research, we propose the method of identifying Aedes mosquito larvae with a digital amplification of 8-12 times without using any microscope lenses attached, using ResNet50 CNN. The proposed model will identity the mosquito larvae by their genus ‘Aedes’ or ‘non- Aedes’ using a digital photo taken by a smartphone or camera in the upside of the larvae body. Hence it would help Health Inspectors, even the general public on identifying Aedes Larvae more efficiently, accurately and conveniently than the traditional method. This study shows that the trained model can identify images of Aedes and Non-Aedes Larvae separately with an accuracy of 86.65%. Furthermore, with using pre-processing techniques, the accuracy level can be enhanced to 98.76% for magnified images.
Design of an auto disconnecting regulator and a safety switch to prevent domestic gas leakages
(Department of Industrial Management, Faculty of Science, University of Kelaniya, Sri Lanka, 2020) Rajapaksha, R.M. I. U.; Perera, P.S.H.; Nandasena, P.K.D.M.; Gunarathna, P.S.T.K.; Kanishka, P.P.D. Gihan; Ranaweera, A.L.A.K.; Kalingamudali, S.R.D.
There is a growing demand for research in various aspects of smart homes. Automated security systems are an integral part of smart homes. Liquid Petroleum Gas (LPG) is one of the popular fuels used in domestic cooking. Therefore, there is a very high demand for LPG fire security systems. In this study, an automated LPG fire security system for domestic gas leakages has been designed and a prototype model is constructed. The designed system automatically takes preventive measures in case of gas leakage. It includes a newly designed automatically disconnecting regulator from the cylinder which shuts OFF gas supply from commercially available gas cylinders, a control circuit for switching OFF the power supply of nearby area of gas leakage and transceiver unit for sending SMS to the corresponding people. It has been designed to operate automatically when LPG concentration reaches to 200 PPM, a value well below the LPG gas inflammable concentration. LPG concentration is sensed by the MQ-5 gas sensor and fed into the microcontroller. The commercially available gas regulator is modified by attaching a spring and solenoid valve. The spring is compressed when the regulator is ON. Once an LPG leakage of appropriate PPM is detected, a pulse is sent to the solenoid valve such that the attached spring gets rest by removing the regulator from the cylinder. At the same time, a warning message will automatically send to the corresponding users and security personnel through a GSM module attached to the circuit. The circuit is embedded with a rechargeable battery to work even in power outage. Additional gas sensors are installed in electrical switches located near to the gas cylinder and kept in connection with the microcontroller through the Bluetooth module to cutoff electrical power to prevent any spark.
Detecting human emotions on Facebook comments
(Department of Industrial Management, Faculty of Science, University of Kelaniya, Sri Lanka, 2020) Chathumali, E.J.A.P.C.; Thelijjagoda, Samantha
Human emotion detection plays a vital role in interpersonal relationships. From the early eras, automatic recognition of emotions has been an active research topic. Today, sharing emotions on social media is one of the most popular activities among internet users. However, when it comes to a specific domain like emotion detection in social media, it is still on a research-level. There are less number of applications have been developed to detect emotions online, using online comments and user comments. The aim of this research is to develop a system that identifies human emotions on Facebook comments. Among the different social media platforms, this research specifically focuses on Facebook comments written in the English language to narrow down the problem. The research is based on Semantic analysis, which comes under Natural Language Processing (NLP) and the system development consists of four major steps, including the extraction of Facebook comments via Graph API, preprocessing, classification and emotion detection. To classify the emotions, a classification model was created by using Naïve Bayes Algorithm. When it comes to marketing, emotions are what lead your onlookers to purchase. By using the detected emotions, marketers can promote their campaigns by changing online advertisements dynamically. The results obtained through testing the system show that it is capable of accurately identifying human emotions hidden in Facebook comments with an accuracy level of 80%, making it highly useful for marketing purposes.
Ecosystem mapping technology to understand innovation challenges in MSME Sector: An analysis of the handicraft sector in Sri Lanka
(Department of Industrial Management, Faculty of Science, University of Kelaniya, Sri Lanka, 2020) Chathura Rajapakse
This paper discusses an application of ecosystem mapping technology to visualize the innovation ecosystem in the Micro, Small and Medium Enterprises (MSME) sector. Ideally, actors in the MSME sector should function as a system, exchanging knowledge, information and other resources, to uplift innovation and create economic value to a country or a region. Visualizing this system enables to understand the deficiencies in the exchange of resources among actors; the deficiencies, which makes innovation and growth challenging to MSMEs. Based on this idea, the study presented in this paper focuses on the handicrafts sector in Sri Lanka and attempt to map its ecosystem based on the manually collected publicly available data on social media. The subsequent analysis of the support ecosystem map reveals a major deficit in the exchange of information technology resources, which hinders the digital innovation and transformation of the MSMEs in the handicrafts sector. The paper further discusses the implications and future research directions.
Extending use-case point-based software effort estimation for Open Source freelance software development
(Department of Industrial Management, Faculty of Science, University of Kelaniya, Sri Lanka, 2020) Senevirathne, Dharshitha Srimal; Wijayasiriwardhane, Thareendhra Keerthi
Accurately predicting the software development effort is very crucial when delivering the software systems on time, within the budget and with the required functionality. Overestimation of the software development effort can lead to losing the projects whereas underestimation can cause budget and schedule overruns. The development effort of a software project depends on various factors and these effort factors associated with the freelance software development are different from those of traditional software development. Software development companies employ various proprietary tools in their projects for their planning, development, testing, etc. However, freelance software developers functioning under tight budgetary constraints are not in a position to afford them. As a result, they tend to use free and open-source tools for their software developments. There are various types of software effort estimation models proposed, published and practiced in the industry. However, there is no such software effort estimation model specifically proposed to estimate the effort of freelance software development. The main objective of this paper is to extend Use Case Point-based software effort estimation for the open-source freelance software development. Initially, details of open source software projects were collected from several freelance software developers. Based on the use case diagrams, Use Case Points counts are then calculated for each project. Taking other effort drivers associated with open source freelance software development also into account, we then estimate the effort of each software development. Our aim is to explore the viability of using Use Case Points as the main effort driver in estimating the effort of open source freelance software development.
Grammatical error detection and correction model for Sinhala language sentences
(Department of Industrial Management, Faculty of Science, University of Kelaniya, Sri Lanka, 2020) Pabasara, H.M.U.; Jayalal, S.
As the national language of Sri Lanka, the greater part of the exercises at most of all the services are completed in Sinhala whereas it is imperative to guarantee the spelling and syntactic accuracy to convey the ideal significance from the perspective of automated materials with the unavailability of resources even though there are enough amount of available materials as hard copy and books. With the high multifaceted nature of the language, it sets aside extensive effort to physically edit the substance of a composed setting. The necessity to overcome this problem has risen numerous years back. But with the complexity of grammar rules in morphologically lavish Sinhala language, the accuracy of the grammar checkers developed so far has been contrastingly lower and thus, to overcome the issue a novel hybrid approach has been introduced. Spell checked Sinhala active sentences being pre-processed, separated nouns and verbs were analyzed with the help of a resourceful part-of- speech-tagger and a morphological analyzer and alongside the sentences were sent through a pattern recognition mechanism to identify its sentence pattern. Then a decision tree-based algorithm has been used to evaluate the verb with the “subject” and output feedback about the correctness of the sentence. To train this decision tree, a dataset consisting of 800 records which included information about 25 predefined grammar rules in Sinhala was used. Finally, the error correction was provided using a machine learning algorithm-based sentence guessing model for the three possible tenses. Conducted research results paved the way to identify the sentence pattern, grammar rules and finally, suggest corrections for identified incorrect grammatical sentences with an acceptable accuracy rate of 88.6 percent which concluded that the proposed hybrid approach was an accurate approach for detecting and correcting grammatical mistakes in Sinhala text.
Identification of factors and classifying the accident severity in Colombo - Katunayake expressway, Sri Lanka
(Department of Industrial Management, Faculty of Science, University of Kelaniya, Sri Lanka, 2020) Kushan, M.A.K.; Chandrasekara, N.V.
Sri Lanka’s expressway system was launched in 2011 and now owns three major expressways. Many peoples choose expressways rather than normal ways due to the reasons of time, traffic, easy of driving, etc. According to police reports of highway main traffic police branch, in recent years the number of accidents occurring in expressways is increasing drastically. Nowadays, the rate of accident occurrence in Colombo-Katunayake Expressway is high compared to the other two expressways and there was no previous research has been done in Sri Lanka regarding accidents on ColomboKatunayake expressway. Therefore, the objective of the study was to identify the factors contributing to accidents on the Colombo-Katunayake Expressway and to develop appropriate machine learning models to classify the severity of the accidents. In this study, 704 total accident cases were considered during the period 2013-2019. Chi-square test, logistic regression, and Kruskal–Wallis tests were used to identify the association between the accident severity and other influential variables found from the literature. Finally, seven variables: time category, driver’s age category, vehicle type, the reason for the accident, number of vehicles involved, cause for accident and rainfall were identified as influencing variables to accident severity under 5% level of significance. Naïve Bayes classification algorithm and probabilistic neural network (PNN) were used in the study to forecast accident severity. A random under-sampling technique was used to overcome the class imbalanced problem persists in the data set considered in the study. The final models developed using the Naïve Bayes algorithm and PNN exhibit 72.14% and 74.29% overall classification accuracy respectively. Both aforementioned models can be considered as suitable models to forecast accident severity in the Colombo-Katunayake expressway where the PNN model exhibits slightly higher accuracy. The final models developed by this study can be used to implement safety improvements against traffic accidents in expressways of Sri Lanka.
Impact of metacognition and age group on contemporary video game interface and gameplay design
(Department of Industrial Management, Faculty of Science, University of Kelaniya, Sri Lanka, 2020) Harischandra, Nandunie N.; Jayakody, Lahiru A.; Madusanka, Tiroshan
Metacognition is about “learning about learning” [1]. However, this theory is far more complicated. This allows people to take charge of their own learning. It involves awareness of how they learn, and evaluation of their learning needs, generating strategies to meet these needs and then implementing the strategies. The theory of metacognition can be identified as humans’ survival instinct. Metacognition also thinks about one’s thinking process such as study skills, memory capabilities, and the ability to monitor learning [2], [3]. Metacognition refers to higher-order thinking which involves active control over the cognitive processes engaged in learning. This concept needs to be explicitly taught along with content instruction. Metacognitive knowledge is about our cognitive processes and our understanding of how to regulate those processes to maximize learning. From ancient times humans have developed their metacognitive skills as a survival factor. For the survival in a video game, players need to follow instructions to get an idea about the gameplay [4]. But most of the time people are likely to skip the instructions without going through them even for the first time. This scenario is noticeable in mobile gameplay. This research has done to identify the factors which affect this dilemma. Metacognition level of a person and age are the variables that were considered for this experiment in order to address the following hypotheses; The more metacognitive skills people have, the more they will find it easy to play never-before-played games. Age range affects performance when it comes to playing the never-before-played games without direct instructions. At the end of this experiment, the first hypothesis became correct while the second one wasn’t. Therefore, people were more likely to ignore direct instructions and go through gameplay successfully when they had a higher metacognitive level, and the age group didn’t seem to affect this factor.
Industry 4.0 readiness assessment for apparel industry: A study in the Sri Lankan context
(Department of Industrial Management, Faculty of Science, University of Kelaniya, Sri Lanka, 2020) Lakmali, Erandika; Vidanagamachchi, Kasuni; Nanayakkara, Julian
Sri Lankan apparel industry is the most significant and driving contributor to the country’s economy by constituting a large portion of GDP. In the highly competitive apparel world, manufacturers search solutions for problems such as worker inadequacy while minimizing the human impact. Therefore, there is a need for apparel manufacturers to enhance value chain processes with the latest technologies. Industry 4.0 is the fourth industrial revolution that transforms the physical production into a combined cyber-physical production environment with IoT and decentralized intelligence. It enhances the process functions from new product development to logistics by providing real-time visibility of the production flow. Existing literature mentions the applications of Industry 4.0 in the apparel industry, but these have not addressed the issue of assessing the readiness for its adaptation in the apparel value chain process. Hence this scrutiny proposes a model to assess the current level of readiness of the Sri Lankan apparel industry to adapt Industry 4.0 technologies and practices. The model was developed based on a systematic review of literature with the industry experts' guidance. The factors that determine the readiness for Industry 4.0 within an organizational context were classified under four categories; People, Process, Technology and Data which were defined as readiness dimensions. The proposed model consists of five readiness levels from 0 to 4 namely: Stranger, Beginner, Intermediate, Advanced and Elite. This model enables managers to measure the readiness for adapting of Industry 4.0 in selected apparel value chain processes by using the specified minimum requirements under each dimension and level. The outcome of this study indicates that Sri Lankan apparel industry is in "Intermediate" level in terms of overall readiness with a value of 1.91 in the predefined readiness scale from 0 to 4.
Initiating customer relationship measurement at small and medium enterprises with low computational cost
(Department of Industrial Management, Faculty of Science, University of Kelaniya, Sri Lanka, 2020) Weerakoon, Tharika Chalani; Rathnayaka, Kapila Tharanga
Modern business entities re-engineer all core business and secondary business activities for customer satisfaction, thereby boosting profit margins. In re-engineering efforts, businesses require immense data processing and decision-making on customers and buying patterns. Still, SMEs face challenges when moving on to customer-centric marketing due to the less accumulated data and resistance towards investing in erudite decision-making tools. Hence, this current research study aims to provide a feasible data mining approach for SMEs in customer recognition with low computation complexity. Data accumulation has happened during the introduction of SME’s mobile application to its customers. Hence, the dataset consists of demographic features age, gender, residency region and occupation of each customer. The proposed approach has two phases as follows; the first phase, the customer demographic data with the target variable of purchase value had subjected to data preprocessing. Null values and noise have treated with a binning method. In the second phase, feature engineering had carried out so categorical variables are in numerically manner. Therefore, the binary encoding was used for categorical. Finally, the dimensionality reduction of the processed data had done using Principal Component Analysis (PCA) to extract the most prominent and customer explanatory attributes within the SME. The PCA yielded 10% of a reduction in total explained variance percentage, meaning that the data had compressed. Using KMeans clustering 4 distinctive clusters were extracted. The usage of PCA had leveraged quick clustering and obtained 4 clusters represented the most impactful customers between the age of 0-17 and occupational level 10. With the implementation of PCA, the dataset narrowed down only with the most prominent features that an SME should care of and with this methodology SME can initiate the practising of more efficient customer data analysis using data mining and machine learning.
Intelligent changeover solution for a domestic hybrid power system
(Department of Industrial Management, Faculty of Science, University of Kelaniya, Sri Lanka, 2020) Herath, H.M.R.M.; Tharindu, K.H.S.; Fernando, K.D.M.; Alahakonn, P.M.K.; Hirshan, R.
Electricity plays a major role in the modern world as almost all the equipment used is operated using electricity. Electricity demand of the world is increasing day by day and there should be a proper mechanism to meet the growing demand and to improve the efficiency of the power systems while continuously providing power with less environmental effects. This paper presents a changeover solution for a solargrid hybrid power system that directly focuses on efficiently utilize the power sources by automatically selecting the power source according to the required power demand. Already available automatic power changeover switches in the market are only capable of selecting one source when the other source is not available. They cannot switch the power source considering the power demand. The novelty of this solution is, it can efficiently select the power source considering the apparent power demand of the house.
IoT based animal classification system using convolutional neural network
(Department of Industrial Management, Faculty of Science, University of Kelaniya, Sri Lanka, 2020) Vithakshana, L.G.C.; Samankula, W.G. D.M.
The kingdom “Animalia” is used to represent all living creatures on the planet earth, which is fallen into six categories. The language is the most common factor to divide humans and animals. Numerous classification techniques can be used for classification purposes, and the classification commonly can be done acoustically and visually. The classification systems are playing a considerable role, and bioacoustics monitoring was a significant field of study. Visual classification of animals is done by using either satellite images or established camera images. Nevertheless, due to some circumstances, image processing techniques cannot be applied. Then the acoustical classification techniques are taken place to encounter those problems. Even with acoustical methods, a remote observing method is required due to a few issues. Applying an IoT based acoustic classification system was designed using Convolutional Neural Networks (CNN), which is beneficial for those who are interested in monitoring ecosystems such as animal scientists, zoologists, and environmentalists. The hardware implementation was designed to collect the data from the place it was placed. The audio clips were preprocessed using the Melfrequency Cepstral Coefficient (MFCC). A CNN architecture based on TensorFlow was used for the training process. To train and test the network, 400 sound clips of two seconds, such that 40 per each ten animal species, which were gathered from online libraries and formatted using Audacity, were used. The network was trained by changing the different gradient descent optimizers and eventually obtained the confusion matrices for each. The best result was gained by the AdaDelta, Gradient Descent, and RMSProp optimizers with 91.3% accuracy for each. Among them, AdaDelta had the most stable and increasing learning approach. As a future extension, to improve accuracy, a large number of data will be used.
Keyword extraction from Tweets using NLP tools for collecting relevant news
(Department of Industrial Management, Faculty of Science, University of Kelaniya, Sri Lanka, 2020) Jayasiriwardene, Thiruni D.; Ganegoda, Gamage Upeksha
Keywords play a major role in representing the gist of a document. Therefore, a lot of Natural Language processing tools have been implemented to identify keywords in both structured and unstructured texts. Text that appears in social media platforms such as twitter is mostly unstructured because of the character limitation. Consequently, a lot of short terms and symbols such as emoticons and URLs are included in tweets. Keyword extraction from grammatically ambiguous text is not easy compared to structured text since it is hard to rely on the linguistic features in unstructured texts. But when it comes to news on twitter, it may contain somewhat structured text than informal text does but it depends on the tweeter, the person who posts the tweet. In this paper, a methodology is proposed to extract keywords from a given tweet to retrieve relevant news that has been posted on twitter, for fake news detection. The intention of extracting keywords is to find more related news efficiently and effectively. For this approach, a corpus that contains tweet texts from different domains is built in order to make this approach more generic instead of making it a domainspecific approach. In fact, the Stanford Core NLP tool kit, Wordnet linguistic database and statistical method are used for extracting keywords from a tweet. For the system evaluation, the Turing test which has human intervention is used. The system was able to acquire an accuracy of 67.6% according to the evaluation conducted.