Skip to main content

Über dieses Buch

This book constitutes the refereed proceedings of the Second International Conference, SLAAI-ICAI 2018, held in Moratuwa, Sri Lanka, in December 2018.
The 32 revised full papers presented were carefully reviewed and selected from numerous submissions. The papers are organized in the following topical sections: ​intelligence systems; neural networks; game theory; ontology engineering; natural language processing; agent based system; signal and image processing.



Intelligence Systems


A Preliminary Study on Kinematic Analysis of Human Hand

The human hands represent an intricate engineering, exquisitely to carry out a variety of tasks. Mimicking human hand is the most challenging task due to the complex structure while, considering the tasks performed by hand. Developing hand prosthesis requires modeling of human hand and the study of its motion capabilities. A human hand model presented in this paper resembles anatomical hand, considering its restrictions and similar motions of the human hand. A 3D model of prosthetic hand that is similar to structure of an adult hand is created using CATIA software. The trajectories and motion of hand model are derived by Simscape toolbox in MATLAB environment. The deformation and strain distribution is obtained using ANSYS software to ensure that hand model will be able to withstand the load it is intended to. Thus the model ratifies for the maximum load and the deformation it would experience on the application of load.
C. N. Savithri, R. Kevin, E. Priya

Fuzzy Logic Based Backtesting System

Identification of favorable trading opportunities is crucial in financial markets as it could bring additional increment in profits. Backtesting is one of the process which consists of analyzing the past price movements and predict the best possible trading strategy. Current approaches focus only on exact matching of market values. In order to overcome deficiencies in current approaches, this research proposes a fuzzy logic based approximate matching approach by using the technical indicators and trading rules. The evaluation results demonstrate that finding approximate matching places for a particular trading strategy has a positive contribution to successful trading and an average trader can be successful in trading buy following those fuzzy logic based trading strategies.
Erandi Praboda, Thushari Silva

Six-State Continuous Processing Model for a New Theory of Computing

In contrast to the computation in Von-Neumann Architecture, the human mind executes processing in the brain by improving speed and accuracy over execution cycles. Further, it has been postulated that the memory is a result of continuous processing and is not separated from processing in the human mind. Similar to that mind model, a six-state continuous processing model with the states New, Ready, Running, Blocked, Sleep and Terminate, has been proposed to implement a special compiler which consists of conditionally evolving memory that aids in improving performance in conventional computation by exploiting 24-Casual Relations explained in Buddhist Theory of Mind. The experiments have been conducted to demonstrate how the proposed computing model increases the performance of execution of source codes and compilers. The result shows a clear increase of performance in computation by avoiding overloading the memory, and ensuring the execution of high quality code segments at the right time.
Chinthanie Weerakoon, Asoka Karunananda, Naomal Dias

Locating the Position of a Cell Phone User Using GSM Signals

Mobile devices are increasingly popular today and mobile location-based services are considered as a profitable opportunity for service providers. Position tracking is essential in implementing many of the new location based services in cellular networks. Mobile positioning technology has become an important area of research as well for emergency and commercial services. There are different ways to track a precise location of a mobile phone or a user. The most widely spread method is the use of built-in GPS module, or the cell tower triangulation. However, most people do not use GPS location service all the time and as a result cellular mobile network based mobile positioning has been an alternative method for tracking. Moreover, these methods have many problems such as low accuracy, high equipment cost, rapid cell-site modifications, and the need of advanced infrastructure. Therefore, a requirement is there to develop a cost effective, accurate, mobile positioning method to replace the current low accuracy and costly methods which use cellular mobile network. In this research project, an Android application was developed to fetch location parameters of connected nearest base stations of a cell phone to locate the coordinates of each and every base station using Google geo-location API. There, combinations of three base station coordinates were used to calculate the approximate location using Pearson Correlation Coefficient approach. Approximated locations were optimized using a Genetic Algorithm to locate final estimated location. This paper discusses the proposed system and the implementation of the proof of concept.
Shazir Shafeeque, G. S. N. Meedin, H. U. W. Ratnayake

Neural Networks


Modeling of Hidden Layer Architecture in Multilayer Artificial Neural Networks

A generated solution for an artificial neural network (ANN) may result in complex computations of neural networks, deployment, and usage of trained networks due to its inappropriate architecture. Therefore, modeling the hidden layer architecture of artificial neural networks remains as a research challenge. This paper presents a solution to achieve the hidden layer architecture of artificial neural networks which is inspired by some facts of neuroplasticity. The proposed method has two phases. First, it determines the number of hidden layers for the best architecture and then removes unnecessary hidden neurons from the network to enhance the performance. Experimental results in several benchmark problems show that the modified network shows better generalization than the original network.
Mihirini Wagarachchi, Asoka Karunananda

A Novel Hybrid Back Propagation Neural Network Approach for Time Series Forecasting Under the Volatility

An Artificial Neural Network (ANN) algorithms have been widely used in machine learning for pattern recognition, classifications and time series forecasting today; especially in financial applications with nonlinear and nonparametric modeling’s. The objective of this study is an attempt to develop a new hybrid forecasting approach based on back propagation neural network (BPN) and Geometric Brownian Motion (GBM) to handle random walk data patterns under the high volatility. The proposed methodology is successfully implemented in the Colombo Stock Exchange (CSE) Sri Lanka, the daily demands of the All Share Price Index (ASPI) price index from April 2009 to March 2017. The performances of the model are evaluated based on the best two forecast horizons of 75% and 85% training samples. According to the empirical results, 85% training samples have given highly accurate in their testing process. Furthermore, the results confirmed that the proposed hybrid methodology always gives the best performances under the high volatility forecasting compared to the separate traditional time series models.
R. M. Kapila Tharanga Rathnayaka, D. M. K. N. Seneviratna

Flood Forecasting Using Artificial Neural Network for Kalu Ganga

Floods are among the natural disasters that cause human hardship and economic loss. Establishing a viable flood forecasting and warning system for communities at risk can mitigate these adverse effects. However, establishing an accurate flood forecasting system is still challenging due to the lack of knowledge about the effective variables in forecasting. The present study has indicated that the use of artificial intelligence, especially artificial neural networks is suitable for flood forecasting systems and identify the input variables, feed them to the ANN and train the model. Then test the model using test data and predict flood level in Kaluthara and Ratnapura area using that ANN model.
Dhananjali Gamage, Kalani Ilmini

Role of Deep Neural Network in Speech Enhancement: A Review

This paper presents a review on different methodologies adopted in speech enhancement and the role of Deep Neural Networks (DNN) in enhancement of speech. Mostly, a speech signal is distorted by background noise, environmental noise and reverberations. To enhance speech, certain processing techniques like Short-Time Fourier Transform, Short-time Autocorrelation and Short-time energy can be adopted. Features such as Logarithmic Power Spectrum (LPS), Mel-Frequency Cepstral Coefficients (MFCC) and Gammatone Frequency Cepstral Coefficient (GFCC) can be extracted and given to DNN for noise classification, so that the noise in the speech can be eliminated. DNN plays a major role in speech enhancement by creating a model with a large amount of training data and the performance of the enhanced speech is evaluated using certain performance metrics.
D. Hepsiba, Judith Justin

Intelligent Time of Use Deciding System for a Melody to Provide a Better Listening Experience

Understanding a melody or a song is quite a difficult task for any machine. This research proposes to analyze notations of the music melodies and to decide the best time to play, sing or listen for any given melody by using the knowledge of Hindustani music trained in an Artificial Neural Network. In the proposed system, pre-process module identifies the Aroha, Awaroha, Vadi and Sanwadi Swara of the melody. Those characteristics that are identified from the pre-process module are input to the Artificial Neural Network (ANN). The system uses the expert knowledge of the Hindustani Raagadari music to train the ANN designed and developed using Tensorflow deep learning platform. Training data set for the learning process has been of size 450 whereas testing data set has been 44 from the total of 494 Raaga details. Trained ANN could achieve a testing accuracy of 84%.
M. W. Sohan Janaka, H. U. W. Ratnayake, I. A. Premaratne

Game Theory


Invoke Artificial Intelligence and Machine Learning for Strategic-Level Games and Interactive Simulations

Computer games are an important sector of the digital economy, computer and entertainment industry are very sophisticated in many ways in the current context of technology. They’ve gone beyond entertainment needs, and the computer game paradigm and technology together are now increasingly used in education, training, storytelling, and wherever it’s necessary to create an appealing and engaging environment. More realism in virtual and artificial environments and more real interfaces to the users can be considered as two main advantages that we get using these techniques. Instead of pre-defined hard coded scripts driving these environments, we will be able to create just the environment and its relative mechanics, so the artificial intelligence could introduce tailored challenges and scenarios to the environment. This paper proposes a Behavioral Driven Procedural Content Generation methodology together with Ternary Neural Networks to be used in interactive strategy-based simulations for effective decision making. This is vital because current approaches like Experience Driven Procedural Content Generation algorithms can be very flexible and one small change could trigger complex changes in the system. Using another model created by using player behavior will be specifying the clamping conditions so the AI is capable of stabilizing itself.
Nishan Chathuranga Wickramarathna, Gamage Upeksha Ganegoda

Ontology Engineering


An Ontological Approach for Knowledge Representation of Dental Extraction Forceps

Tooth extraction is a common surgical procedure in the dental field. If the procedure is carried out without proper knowledge about the tooth and the extracting instruments, it may lead to complexity on extraction procedure or even cause damages to the patients’ jaws. Therefore, the information and knowledge must be provided in a structured and complete way and in a context specific manner. Initially, information regarding the dental extraction forceps were gathered from the experts in the field, considering the importance of sharing the knowledge on dental extraction. Subsequently, an ontology on dental extraction forceps was developed. Next, the developed ontology was evaluated by ontology experts in an iterative approach. Finally, a knowledge sharing portal was developed and validated. We strongly believe that our dental extraction knowledge sharing portal can support the dental students, dentists as well as their assistants to improve the sharing of knowledge and learning practices.
Shanmuganathan Vasanthapriyan

Natural Language Processing


Digital Assistant for Supporting Bank Customer Service

Digital Assistants are trending in most industries. Despite the fact that it invaded the banking industry a few years ago, the concept is still new in the Sri Lankan banking context. A Chat-bot improves customer satisfaction providing solutions to customers, using the most preferred user interaction method called chat application as a solution to the usually inefficient, time-consuming processes which are followed in call centres. The proposed system is a question answering system which facilitates customers to solve their day-to-day banking related questions. The system provides answers to a set of most Frequently Asked Questions (FAQ) related to the banking domain. Customer inquiries are extracted and converted using natural language processing techniques understood by the system. It formulates and presents an appropriate answer using a set of preconfigured templates referring to the answer saved in a knowledge base. Knowledge is represented in the form of an Ontology.
Dinithi Weerabahu, Agra Gamage, Chathurya Dulakshi, Gamage Upeksha Ganegoda, Thanuja Sandanayake

A Novel Dialogue Manager Model for Spoken Dialogue Systems Based on User Input Learning

The complexity of the dialogue manager is a major issue in spoken dialogue systems. In this work, a novel dialogue manager based on user input learning is proposed to overcome this issue. In the proposed model back-end functionality is considered as a set of functions a user can trigger through the dialogue manager. It uses these functions as classes for the classification of user inputs. To maintain the context of the dialogue interactions, a context tree is used. Consequently, the model performs its task as two classification tasks to identify the function a user input may trigger and use the context to maintain the discourse of the dialogue. The model shows promising results and proves that a dialogue manager can be integrated into a spoken dialogue system much more directly with less hassle.
M. F. Ahmed Shariff, Ruwan D. Nawarathna

Text Mining-Based Human Computer Interaction Approach for On-line Purchasing

E-commerce websites have created a great opportunity not only for businesses but also for consumers to perform their transactions directly. These transactions can be classified into different types such as consumer-oriented factors, behavioral factors, and Human Computer Interaction (HCI)-based factors. This study uses a twofold approach. In the first phase, prominent HCI factors are identified through existing literature namely; accessibility, simplicity, and usefulness which enhances the interaction of people with E-commerce websites. In the second phase of the study, we conducted a detailed experiment by varying the identified HCI factors towards consumer interaction using text-mining approach. Various approaches have been utilized to identify the relationship between factors affecting the consumers’ online purchasing behavior. Most of the cases, those studies focused on one website to identify the HCI factors pertaining to it. To overcome the research gap in current literature, the authors have built a novel and a unique approach to assess the factors related to HCI in enhancing online purchasing experience of diverse customer settings.
Nadeeka Malkanthi, Thashika D. Rupasinghe

Feature Based Opinion Mining for Hotel Profiling

Hotel profiling plays an important role in hotel recommendation. With the proliferation of huge amount of user-generated-reviews on web-sites, hotel profiling has become more challenging as these reviews and embedded opinions could indirectly drive hotels. Comprehensive hotel profiling based on review analysis could help people to get an overall opinion on hotels and hence to facilitate mindful tourism. To avoid deficiencies of many other recent researches, this research focuses more on the feature-based opining mining rather analysing only sentiments of the reviews. Thus, a semantic profiling approach which integrates a machine learning technique, part-of-speech (PoS) tagging and Ontology is proposed for feature-based hotel profiling. PoS tagging is used for recognising patterns of opinions and SentiWordNet is used to resolve semantic heterogeneity of the opinion phrases and to classify them. Feature-based analysis could generate the feature level opinion about a hotel in several aspects including food, hospitality and environment.
Dilum Gunathilaka, Shamila Pathirana, Sasanka Senarathne, Jithmi Weerasekara, Thushari Silva

Agent Based System


A Hybrid Agent System to Detect Stress Using Emotions and Social Media Data to Provide Coping Methodologies

Final year undergraduates in Sri Lanka are more likely to experience high levels of stress due to the high competition in the education system. Living with high levels of stress has the possibility of putting a person’s entire well-being at a great risk. Today more and more students are suffering from various levels of stress. Too much stress will bring a variety of physical and psychological problems including anxiety, depression and even suicide to growing youths. Traditional face-to-face stress detection and relief methods do not work, confronted with undergraduates who are reluctant to express their negative emotions to the people in real life. In this paper, the authors present undergraduates-oriented intelligent chatting system which aims to act as a virtual friend to listen, understand, comfort, encourage, and guide stressful undergraduates to pour out their bad feelings and thus release the stress by suggesting stress coping mechanisms to follow and to be guided. Our user study demonstrates that this system is effective on sensing and coping with undergraduates stress.
Ridmal Liyanagamage, Shakina Kitchilan, Roshan Maddumage, Shazeeka Kitchilan, Nishantha Kumarasinghe, Subha Fernando

Thinking Like Humans: A New Approach to Machine Translation

Existing machine translation approaches do not adequately mimic how humans do translation from one natural language to another. This paper presents a novel approach to machine translation that is inspired by how humans translate natural languages. We have exploited the theory of psycholinguistic sentence-parsing to develop a human-like machine translation system. This approach has been modeled as a multi-agent system, named EnSiMaS, which translates an English sentence into Sinhala sentence. The multi-agent system has been implemented through the MaSMT framework with two manager agents and over 100 agents which deliberate on different aspects of machine translation. These agents are clustered into eight-agent swarms to consider morphological, syntactic, and semantic concerns of the source and the target languages. The EnSiMaS system has been tested with the different types of sentences and successful results were obtained.
Budditha Hettige, Asoka Karunananda, Gorge Rzevski

Rice Express: A Communication Platform for Rice Production Industry

Rice production can consider as the main production area in the agriculture industry. Because of the poor communication among farmers, buyers and transporters Sri Lanka is a high-cost rice producer. Rice production cost can significantly reduce through the communication between relevant persons in right time. Thus Multi-Agent technology can be used to handle the communication productively. This paper presents a multi-agent solution for the agriculture industry, namely, Rice Express, which is capable of communicating between the persons in the rice production industry. Rice Express has been developed through the MaSMT framework. The system comprises three types of agents, namely, farmers, buyers and transporters. With the relevant agent communication among farmers, buyers, and transporters, the system should be capable of reducing transport cost significantly. The Rice Express has been successfully tested in the practical environment, and successful results were obtained.
M. A. S. T. Goonatilleke, M. W. G. Jayampath, B. Hettige

Signal and Image Processing


Diagnosis of Coronary Artery Diseases and Carotid Atherosclerosis Using Intravascular Ultrasound Images

Cardiovascular diseases are of paramount importance as large number of deaths is caused, if not diagnosed and treated at the right time. Ultrasound examination complements other imaging modalities such as radiography, and allows more definite diagnostic tests to be conducted. This modality is non-invasive in nature, widely used in diagnosis of cardiovascular diseases. Recently, two leading ultrasound based techniques are used for the assessment of atherosclerosis: B-mode ultrasound used in measurement of carotid artery intima thickness and intravascular ultrasound. These techniques provide images in real time, portable, substantially lower in cost and no harmful ionizing radiations are used in imaging. The processing of ultrasound image takes a major role in the accurate diagnosis of the disease level. The diagnostic accuracy depends on the time to read the image and the experience of the practitioner to interpret the correct information. Computer aided methods for the analysis of the intravascular ultrasound images can assist in better measurement of plaque deposition in the coronary artery. In this paper, the level of plaque deposition is identified using Otsu’s segmentation method and classification of plaque deposition level is performed using Back Propagation Network (BPN) and Support Vector Machine (SVM). The result shows SVM classifies more significantly in comparison with the BPN network.
K. V. Archana, R. Vanithamani

Performance Analysis: Preprocessing of Respiratory Lung Sounds

Computerized lung sound analysis for automatic detection and classification of adventitious lung sounds is an emerging technique for the diagnosis of pulmonary diseases. Automated analysis of lung sound signals involves acquisition of clean lung sounds and identification of key factors present in the signal to aid the physician in recognizing the category of adventitious lung sounds. There is a possibility that the acquired lung sounds may be corrupted with interferences such as heart sound, artifacts due to improper mounting of sensor and power line interference. It also depends upon the environment in which the signals are recorded. Therefore preprocessing of the signal plays the key role in diagnosis and interpretation of lung sound. In this work the lung sounds are preprocessed using Recursive Least Mean Square (RLS), Least Mean Square (LMS), Square root Recursive Least Mean Square (SRLS), Discrete Wavelet Transform (DWT) and Total Variation De-noising (TVD) methods. The performance metrics Mean Square Error (MSE), Mean Absolute Error (MAE), Signal to Noise Ratio (SNR), Peak Signal to Noise Ratio (PSNR) and Cross Correlation (CC) are computed for the evaluation of RLS, LMS, SRLS, DWT and TVD. It is observed from the results that the DWT performs better compared to RLS, LMS, SRLS and TVD in removing the artifacts from the lung sounds.
G. Shanthakumari, E. Priya

A Classification Based Approach to Predict the Gender Using Craniofacial Measurements

In archaeological, forensic, clinical and surgical fields, it is important to decide the gender of an individual and to know the facial asymmetry. The objective of this study is to ascertain the asymmetry of the skull and to predict the gender based on the shape and dimensions of the infraorbital foramina (IOF) and its position in relation to maxillary teeth, supraorbital foramen (SOF) and clinically relevant anatomical landmarks. Linear discriminant analysis (LDA), binary logistic regression (BLR), support vector machine (SVM) and bagging CART algorithms were used to predict the gender and results were validated using 10-fold cross-validation technique. A significant variation on the position of IOF in relation to anatomical landmarks was observed and the relative position of the IOF varies between the genders. Moreover, the left side measurements were larger than the right-side measurements and hence human skull is asymmetric. The outcome of the study proposes the classification-based approach to predict the gender of the individuals. The Bagging CART model performed well in terms of both accuracy and precision when compared with other methods.
Maneesha M. M. Arachchi, Lakshika S. Nawarathna, Roshan Peiris, Deepthi Nanayakkara

An Optimized Predictive Coding Algorithm for Medical Image Compression

This article proposes a novel algorithm which helps in efficient transmission and storage of medical images. The conventional prediction algorithm is modified in such a way that they provide high compression without any further degradation in the image quality. A binary mask is generated based on the optimized threshold value for the image data. Then prediction is done for the masked coefficients to eliminate high error values caused by lower range of coefficients. An appropriate prediction function which gives less entropy for the input image is selected and encoded. The experimental results showed a maximum of 45% improvement in compression ratio compared to the normal prediction process. The proposed modified prediction algorithm can efficiently replace the prediction step in any lossy or lossless compression algorithms. They can also be utilized as a part of compression in any contextual compression techniques. Any kind of transformation approach can be used in hybridization with this proposed optimized prediction model to perform better.
J. Anitha, P. Eben Sophia, D. Jude Hemanth

Palm Vein Recognition Based on Competitive Code, LBP and DCA Fusion Strategy

Information fusion mainly includes feature level fusion, matching-score level fusion and decision level fusion. However, the feature level fusion is considered to be a more effective fusion method because of the more biometric data than the matching fraction fusion and the decision level fusion. Feature level fusion is the extraction of feature information from the source image and the multiple features are analyzed, processed and integrated to get a single fusion image feature. In this paper, the palm vein features are extracted with Competition code and local binary pattern (LBP), respectively, to obtain two different palm vein features. Then two features are fused using discriminant correlation analysis (DCA). DCA is a feature level fusion technology, which associates class associations to the correlation analysis of feature sets. DCA implements effective feature fusion by maximizing the pairwise correlation between the two feature sets, and eliminating inter class correlation and restricting the correlation within the class. The Competition code uses the directional characteristics of the image to extract the palm vein features, while the LBP is an operator used to describe the local texture features of the image. The two features are complementary. Using DCA to combine the two characteristics of the palm vein can achieve a good classification effect. This paper uses the multispectral near-infrared palm vein image database of Hong Kong Polytech University for testing. Compared with the single palm vein Competition code feature or LBP feature, the DCA combines the two characteristics of the palm vein, which not only shortens the classification time in some degree, but also improves the recognition rate to 99.8% in the case of training samples of 9 and test samples of 3.
Xiyu Wang, Hengjian Li


Weitere Informationen

Premium Partner