Search

Gated Recurrent Neural Networks (RNNs) such as LSTM and GRU have been highly effective in handling sequential time series data in recent years. Although Gated RNNs have an inherent ability to learn complex temporal dynamics, there is potential for further enhancement by enabling these deep learning networks to directly use time information to recognise time-dependent patterns in data and identify important segments of time. Synonymous with time series data in real-world applications are missing values, which often reduce a model’s ability to perform predictive tasks. Historically, missing values have been handled by simple or complex imputation techniques as well as machine learning models, which manage the missing values in the prediction layers. However, these methods do not attempt to identify the significance of data segments and therefore are susceptible to poor imputation values or model degradation from high missing value rates. This paper develops Cyclic Gate enhanced recurrent neural networks with learnt waveform parameters to automatically identify important data segments within a time series and neglect unimportant segments. By using the proposed networks, the negative impact of missing data on model performance is mitigated through the addition of customised cyclic opening and closing gate operations. Cyclic Gate Recurrent Neural Networks are tested on several sequential time series datasets for classification performance. For long sequence datasets with high rates of missing values, Cyclic Gate enhanced RNN models achieve higher performance metrics than standard gated recurrent neural network models, conventional non-neural network machine learning algorithms and current state of the art RNN cell variants.

Extreme learning machines (ELMs) has been theoretically and experimentally proved to achieve promising performance at a fast learning speed for supervised classification tasks. However, it does not perform well on imbalanced binary classification tasks and tends to get biased toward the majority class. Besides, since a large amount of training data with labels are not always available in the real world, there is an urgent demand to develop an efficient semi-supervised version of ELM for imbalanced binary classification tasks. In this article, owing to the distinct insensitivity of area under the ROC curve (AUC) to both class skews and changes of class distributions, we focus the study on integrating AUC maximization into the ELM framework to tackle with imbalanced binary classification tasks well. By demystifying the AUC metric with the ELM framework, we develop a new AUC-based ELM called AUC-ELM for imbalanced binary classification, which essentially is revealed to be equivalent to an ELM on another transformed data space. Accordingly, its semi-supervised version called SAUC-ELM is also developed. Both AUC-ELM and SAUC-ELM have the distinctive merits: 1) they share the advantage of ELM in both generalization capability and training efficiency, and further uniquely tailored for imbalanced binary classification tasks and 2) in contrast to the existing imbalanced variants of ELM, such as class-specific cost regulation ELM and semi-supervised ELM, they have fewer parameters to tune, thereby reducing the computational cost for model selection. Experiments on a heap of datasets show that both AUC-ELM and SAUC-ELM outperform the other comparative methods in terms of both classification performance and training speed.

The use of sensors for monitoring livestock has opened up new possibilities for the management of livestock in extensive grazing systems. The work presented in this paper aimed to develop a model for predicting the metabolisable energy intake (MEI) of sheep by using temperature, pitch angle, roll angle, distance, speed, and grazing time data obtained directly from wearable sensors on the sheep. A Deep Belief Network (DBN) algorithm was used to predict MEI, which to our knowledge, has not been attempted previously. The results demonstrated that the DBN method could predict the MEI for sheep using sensor data alone. The mean square error (MSE) values of 4.46 and 20.65 have been achieved using the DBN model for training and testing datasets, respectively. We also evaluated the influential sensor data variables, i.e., distance and pitch angle, for predicting the MEI. Our study demonstrates that the application of machine learning techniques directly to on-animal sensor data presents a substantial opportunity to interpret biological interactions in grazing systems directly from sensor data. We expect that further development and refinement of this technology will catalyse a step-change in extensive livestock management, as wearable sensors become widely used by livestock producers.

Bio-waste valorization of Mango Seed (MS) husk to Activated Carbon (AC) was conducted via carbonization followed by chemical activation using KOH for sustainable supercapacitor material. Various physicochemical techniques were used to characterise both MS husk and AC, while electrochemical techniques were used to assess the energy storage capability. Carbonized MS husk was activated at four different temperatures 800, 900, 1000, and 1100 °C. Optimal activation temperature was found to be 1100 °C resulting in a hierarchical porous structure, possessing the highest specific surface area 1943 m2 g−1 and average pore volume 0.397 cm3 g−1, the critical parameters for their energy storage performance. In a three-electrode configuration, the MS-AC sample synthesized at 1100 °C (MS-AC 1100) exhibited a maximum specific capacitance of 135 F g−1 at 5 mA cm−2 with the energy density of 19 Wh kg−1 at the power density of 1077 W kg−1, which is higher than the commercially available AC. The MS-AC 1100 delivered 97% of the capacitance when increasing the current densities from 2 to 10 mA cm−2. The capacitance retention is 100% at 5 mA cm−2 with 1000 cycles. Moreover, Machine Learning (ML) models are used to identify the non-linear pattern for predicting the AC capacitance using various input parameters. Out of the four ML models considered in this work, the Multi-Layer Perceptron (MLP) model outperformed the other models to better correlate the predicted data and the experimental measurements. This study demonstrates the economic perspective of repurposing MS husk bio-waste to a low-cost super-capacitor electrode material and developing high-performance carbon materials to aid the integration of renewable energy into electricity systems.

Analysis of time series data has been a challenging research subject for decades. Email traffic has recently been modelled as a time series function using a Recurrent Neural Network (RNN) and RNNs were shown to provide higher prediction accuracy than previous probabilistic models from the literature. Given the exponential rise of email workloads which need to be handled by email servers, in this paper we first present and discuss the literature on modelling email traffic. We then explain the advantages and limitations of different approaches as well as their points of agreement and disagreement. Finally, we present a comprehensive comparison between the performance of RNN and Long Short Term Memory (LSTM) models. Our experimental results demonstrate that both approaches can achieve high accuracy over four large datasets acquired from different universities’ servers, outperforming existing work, and show that the use of LSTM and RNN is very promising for modelling email traffic.

Feature Selection (FS) methods have been studied extensively in the literature, and there are a crucial component in machine learning techniques. However, unsupervised text feature selection has not been well studied in document clustering problems. Feature selection could be modelled as an optimization problem due to the large number of possible solutions that might be valid. In this paper, a memetic method that combines Differential Evolution (DE) with Simulated Annealing (SA) for unsupervised FS was proposed. Due to the use of only two values indicating the existence or absence of the feature, a binary version of differential evolution is used. A dichotomous DE was used for the purpose of the binary version, and the proposed method is named Dichotomous Differential Evolution Simulated Annealing (DDESA). This method uses dichotomous mutation instead of using the standard mutation DE to be more effective for binary purposes. The Mean Absolute Distance (MAD) filter was used as the feature subset internal evaluation measure in this paper. The proposed method was compared with other state-of-the-art methods including the standard DE combined with SA, which is named DESA in this paper, using five benchmark datasets. The F-micro, F-macro (F-scores) and Average Distance of Document to Cluster (ADDC) measures were utilized as the evaluation measures. The Reduction Rate (RR) was also used as an evaluation measure. Test results showed that the proposed DDESA outperformed the other tested methods in performing the unsupervised text feature selection.

In this paper, a model for controlling paddy drying by incorporating accumulated temperature is introduced and defined. Drying experiments using freshly harvested paddy were conducted at different levels of drying air parameters including temperature (T = 27°C, 31°C, 35°C, 39°C, and 43°C), relative humidity (RH = 45%, 50%, 55%, 60%, and 65%), initial moisture content (M0 = 17%, 19%, 21%, 23%, and 25%) and airspeed (V = 0.4, 0.5, 0.6, 0.7, and 0.8 m s−1). When T = 31°C, RH = 60%, M0 = 19%, and V = 0.5 ms−1, the crack additional percentage reached a minimum value of 0.508%, with the average precipitation rate at 0.719%·h. Further, an accumulated temperature and quality chart was provided in this paper. The x-coordinate of this chart is temperature, and the initial moisture content represents the y-coordinate. It covers the drying conditions of the actual dryer well and has a wide range of applicability for real-world environment. The model developed in this study not only provides a scientific reference for precise drying and intelligent control of paddy but also guide the actual drying operation.

The effectiveness of Gesture-Based Learning System (GBLS) has been reported in some recent studies. However, not many of those studies have investigated on how GBLS mode influences the learning outcomes. The aim of this study therefore focuses on investigating how GBLS mode impacts the learning outcomes. The findings of this study revealed that GBLS's features positively affect the students' intrinsic motivation. Consequently, the increase in the intrinsic motivation leads to improving the learning outcomes; this study also showed that GBLS's features indirectly influence the learning outcomes via intrinsic motivation. In other words, this study found that the GBLS's features (interactivity and multimodality) create an instructional learning environment that positively influences the students' intrinsic motivation. The increase of the students' positive intrinsic motivation led to enhancing the learning achievements of students.

Nave Bayes, k-nearest neighbors, Adaboost, support vector machines and neural networks are five among others commonly used text classifiers. Evaluation of these classifiers involves a variety of factors to be considered including benchmark used, feature selections, parameter settings of algorithms, and the measurement criteria employed. Researchers have demonstrated that some algorithms outperform others on some corpus, however, inconsistency of human labeling and high dimensionality of feature spaces are two issues to be addressed in text categorization. This paper focuses on evaluating the five commonly used text classifiers by using an automatically generated text document collection which is labeled by a group of experts to alleviate subjectivity of human category assignments, and at the same time to examine the influence of the number of features on the performance of the algorithms.

Digital pathology has attracted significant attention in recent years. Analysis of Whole Slide Images (WSIs) is challenging because they are very large, i.e., of Giga-pixel resolution. Identifying Regions of Interest (ROIs) is the first step for pathologists to analyse further the regions of diagnostic interest for cancer detection and other anomalies. In this paper, we investigate the use of RCNN, which is a deep machine learning technique, for detecting such ROIs only using a small number of labelled WSIs for training. For experimentation, we used real WSIs from a public hospital pathology service in Western Australia. We used 60 WSIs for training the RCNN model and another 12 WSIs for testing. The model was further tested on a new set of unseen WSIs. The results show that RCNN can be effectively used for ROI detection from WSIs.

Recurrent Neural Networks (RNN), Long Short-Term Memory Networks (LSTM), and Memory Networks which contain memory are popularly used to learn patterns in sequential data. Sequential data has long sequences that hold relationships. RNN can handle long sequences but suffers from the vanishing and exploding gradient problems. While LSTM and other memory networks address this problem, they are not capable of handling long sequences (50 or more data points long sequence patterns). Language modelling requiring learning from longer sequences are affected by the need for more information in memory. This paper introduces Long Term Memory network (LTM), which can tackle the exploding and vanishing gradient problems and handles long sequences without forgetting. LTM is designed to scale data in the memory and gives a higher weight to the input in the sequence. LTM avoid overfitting by scaling the cell state after achieving the optimal results. The LTM is tested on Penn treebank dataset, and Text8 dataset and LTM achieves test perplexities of 83 and 82 respectively. 650 LTM cells achieved a test perplexity of 67 for Penn treebank, and 600 cells achieved a test perplexity of 77 for Text8. LTM achieves state of the art results by only using ten hidden LTM cells for both datasets.

Data mining is the analytics and knowledge discovery process of analyzing large volumes of data from various sources and transforming the data into useful information. Various disciplines have contributed to its development and is becoming increasingly important in the scientific and industrial world. This article presents a review of data mining techniques and applications from 1996 to 2016. Techniques are divided into two main categories: predictive methods and descriptive methods. Due to the huge number of publications available on this topic, only a selected number are used in this review to highlight the developments of the past 20 years. Applications are included to provide some insights into how each data mining technique has evolved over the last two decades. Recent research trends focus more on large data sets and big data. Recently there have also been more applications in area of health informatics with the advent of newer algorithms.

The pervasiveness of Social Media and user-generated content has triggered an exponential increase in global data volumes. However, due to collection and extraction challenges, data in many feeds, embedded comments, reviews and testimonials are inaccessible as a generic data source. This paper incorporates Knowledge Management framework as a paradigm for knowledge management and data value extraction. This framework embodies solutions to unlock the potential of UGC as a rich, real-time data source for analytical applications. The contributions described in this paper are threefold. Firstly, a method for automatically navigating pagination systems to expose UGC for collection is presented. This is evaluated using browser emulation integrated with dynamic data collection. Secondly, a new method for collecting social data without any a priori knowledge of the sites is introduced. Finally, a new testbed is developed to reflect the current state of internet sites and shared publicly to encourage future research. The discussion benchmarks the new algorithm alongside existing data extraction techniques and provides evidence of the increased amount of UGC data made accessible by the new algorithm.

In the special issue on computational intelligence (CI) and complex systems a plethora of approaches is presented on how to apply computation intelligence in modeling, control, decision support, and optimization of complex problems. The three main components of CI are fuzzy systems, evolutionary and population based algorithms, and artificial neural networks. In addition, there are other related techniques, among others, chaos theory and subjective probability. Often, the combinations and hybrids of these methods, sometimes complemented by classic mathematical or modeling tools, turn out to be most efficient in the solution of real life problems. In the series of 26 papers, illustrations for almost all of these methods may be found...

Augmented Reality (AR) is a technology that augments reality with either two or three dimensional computer generated imagery (CGI), objects and/or information, and allows users to interact with them. AR on mobile devices are evolving and offer a great deal of potential in terms of learning and training. This paper discusses the development process of a mobile prototype learning environment that utilises mobile-Augmented Reality (mAR). The prototype is called the Human Anatomy in Mobile Augmented Reality or HuMAR, and the selected learning topic is the anatomy of the human skeletal structure. The main objective of HuMAR is to aid students and it could potentially enhance their learning process. There has been a report stating that there is a decline in retaining and generating long lasting information longer when learning the abovementioned topic. This paper describes the theory, concept, prototype development and results of HuMAR taken from a pilot test. The pilot test used the experimental method with science's students from three different universities. The objectives of the pilot test were to consolidate users’ experience from a didactic and technical point of view. Based on the results of the pilot test, it is concluded that students were satisfied with HuMAR in terms of its usability and features; which in turn could have a positive impact in their learning process.

Visual design plays an important role in grabbing web users’ attention in an online environment. Previous research has demonstrated that different types of visual design causes different impact towards the end-users. This paper observes the impact of persuasive visual towards users’ first impression, attitudes, and behaviours. It extends existing web visual design by empirically examining the critical roles of the principles of social influence in the form of visual persuasion in motivating users to have a favourable impression of a particular website. Survey data was collected in an experimental study that was conducted online. Structural model assessment is carried out using confirmatory factor analysis (CFA) in conjunction with PLS-SEM analyses. The general analysis of model fit indicates that the two models proposed in this paper surpassed the cut off values for model acceptance with most of the model fit criteria reflects outstanding explanatory power. The result of the analysis indicates that visual persuasion has a big impact in influencing users’ attitudes on the web; strong enough to affect their behavioural intention.

This study aims to verify the learning effectiveness of a desktop virtual reality (VR)-based learning environment, and to investigate the effects of desktop VR-based learning environment on learners with different spatial abilities. The learning outcome was measured cognitively through academic performance. A quasi pretest-posttest experimental design was employed for this study. A total of 431 high school students from four randomly selected schools participated in this study where they were randomly assigned to either experimental or control groups based on intact classes. Findings indicate a significant difference in the performance achievement between the two groups with students performed better using desktop virtual reality. A possible explanation is that the desktop virtual reality instructional intervention has helped to reduce extraneous cognitive load and engages learners in active processing of instructional material to increase germane cognitive load. A significant interaction effect was found between the learning mode and spatial ability with regard to the performance achievement. Further analysis shows a significant difference in the performance of low spatial ability learners in the experimental and control groups, but no statistically significant difference in the performance of high spatial learners in both groups. The results signify that low spatial ability learners' performance, compared with high spatial ability learners, appeared to be more positively affected by the desktop VR-based learning environment which is supported by the ability-as-compensator hypothesis, and can be explained by the cognitive load theory. Desktop VR-based learning environment shows positive effect on students' performance.Performance achievement varies as a function of spatial ability and learning mode.Low spatial ability learners benefit more in VR learning mode.

Effective system development to support higher education institutions is important. There are two steps in developing a good system and they are system architecture analysis and design. The first task of this research is to design the use case diagrams, system overview and the system architecture. The second is to evaluate the system architecture of a proposed Student Relationship Management (SRM) system using Internet of Things (IoT) to collect digital footprint of higher education institutions. The outcome of this research include the system architecture of the proposed student relationship management system (SRMS)-IoT consists of six main parts: 1) service stations, 2) system identification, 3) system integration API, 4) SRM internal system, 5) report analytic and 6) web server and database server. Evaluation of the results shows an overall appropriateness at a very high level: the overall appropriateness of usability result was also at a very high level, which shows that this research is appropriate to be used as a guideline for further system development to support student services, and to promote learning and analysis of student behaviours in higher education institutions.

Content filtering in a mobile recommendation system plays a vital role in providing solution to help mobile device users obtain their desire content. However, mobile content recommendation systems have problems and limitations related to cold start and sparsity. These problems can be viewed as a user’s first time connection to a mobile recommendation system and initial rating of the content in an early stage of the system. Hence, to obtain personalized content for mobile user, mobile content filtering is needed. This paper proposes a framework for integrated mobile content recommendation. The framework makes use of classification and adaptive association rule techniques to build an integrated model. The results demonstrate that the proposed framework outperforms related techniques. This can address the problem of sparsity for mobile content recommendation systems. To cite this document: Worapat Paireekreng, Kok Wai Wong, and Chun Che Fung, "A framework for integrated mobile content recommendation", International Journal of Electronic Commerce Studies, Vol.4, No.2, pp. 185-202, 2013. Permanent link to this document: http://dx.doi.org/10.7903/ijecs.1038

As email workloads keep rising, email servers need to handle this explosive growth while offering good quality of service to users. In this work, we focus on modeling the workload of the email servers of four universities (2 from Greece, 1 from the UK, 1 from Australia). We model all types of email traffic, including user and system emails, as well as spam. We initially tested some of the most popular distributions for workload characterization and used statistical tests to evaluate our findings. The significant differences in the prediction accuracy results for the four datasets led us to investigate the use of a Recurrent Neural Network (RNN) as time series modeling to model the server workload, which is a first for such a problem. Our results show that the use of RNN modeling leads in most cases to high modeling accuracy for all four campus email traffic datasets.

This study examined how desktop virtual reality (VR) enhances learning and not merely does desktop VR influence learning. Various relevant constructs and their measurement factors were identified to examine how desktop VR enhances learning and the fit of the hypothesized model was analyzed using structural equation modeling. The results supported the indirect effect of VR features to the learning outcomes, which was mediated by the interaction experience and the learning experience. Learning experience which was individually measured by the psychological factors, that is, presence, motivation, cognitive benefits, control and active learning, and reflective thinking took central stage in affecting the learning outcomes in the desktop VR-based learning environment. The moderating effect of student characteristics such as spatial ability and learning style was also examined. The results show instructional designers and VR software developers how to improve the learning effectiveness and further strengthen their desktop VR-based learning implementation. Through this research, an initial theoretical model of the determinants of learning effectiveness in a desktop VR-based learning environment is contributed.

In most classification problems, sometimes in order to achieve better results, data cleaning is used as a preprocessing technique. The purpose of data cleaning is to remove noise, inconsistent data and errors in the training data. This should enable the use of a better and representative data set to develop a reliable classification model. In most classification models, unclean data could sometime affect the classification accuracies of a model. In this paper, we investigate the use of misclassification analysis for data cleaning. In order to demonstrate our concept, we have used Artificial Neural Network (ANN) as the core computational intelligence technique. We use four benchmark data sets obtained from the University of California Irvine (UCI) machine learning repository to investigate the results from our proposed data cleaning technique. The experimental data sets used in our experiment are binary classification problems, which are German credit data, BUPA liver disorders, Johns Hopkins Ionosphere and Pima Indians Diabetes. The results show that the proposed cleaning technique could be a good alternative to provide some confidence when constructing a classification model.

This paper proposes a non-segmented document clustering method using self-organizing map (SOM) and frequent max substring technique to improve the efficiency of information retrieval. SOM has been widely used for document clustering and is successful in many applications. However, when applying to non-segmented document, the challenge is to identify any interesting pattern efficiently. There are two main phases in the propose method: preprocessing phase and clustering phase. In the preprocessing phase, the frequent max substring technique is first applied to discover the patterns of interest called Frequent Max substrings that are long and frequent substrings, rather than individual words from the non-segmented texts. These discovered patterns are then used as indexing terms. The indexing terms together with their number of occurrences form a document vector. In the clustering phase, SOM is used to generate the document cluster map by using the feature vector of Frequent Max substrings. To demonstrate the proposed technique, experimental studies and comparison results on clustering the Thai text documents, which consist of non-segmented texts, are presented in this paper. The results show that the proposed technique can be used for Thai texts. The document cluster map generated with the method can be used to find the relevant documents more efficiently.

Games technology has undergone tremendous development. In this article, the authors report the rapid advancement that has been observed in the way games software is being developed, as well as in the development of games content using game engines. One area that has gained special attention is modeling the game environment such as terrain and buildings. This article presents the continuous level of detail terrain modeling techniques that can help generate and render realistic terrain in real time. Deployment of characters in the environment is increasingly common. This requires strategies to map scalable behavior characteristics for characters as well. The authors present two important aspects of crowd simulation: the realism of the crowd behavior and the computational overhead involved. A good simulation of crowd behavior requires delicate balance between these aspects. The focus in this article is on human behavior representation for crowd simulation. To enhance the player experience, the authors present the concept of player adaptive entertainment computing, which provides a personalized experience for each individual when interacting with the game. The current state of game development involves using very small percentage (typically 4% to 12%) of CPU time for game artificial intelligence (AI). Future game AI requires developing computational strategies that have little involvement of CPU for online play, while using CPU’s idle capacity when the game is not being played, thereby emphasizing the construction of complex game AI models offline. A framework of such nonconventional game AI models is introduced.

When designing educational computer games, designers usually consider target age, interactivity, interface and other related issues. They rarely explore the genres which should employ into one type of educational game. Recently some digital game-based researchers attempt to combine game genre with learning theory. Different researcher use different pedagogy conceptions. The purpose of this paper is to investigate the appropriate computer genre in designing effective educational computer games that can be used as ;learning tools in Thai undergraduate classrooms by integrating the learning conceptions from the previous work of those researchers. The following steps are used to conduct this study: summarize the survey of Thai students’ opinions toward game genre; analyse the characteristics of genre of computer game; match those characteristics of genre with learning theories; and propose the conceptual model from the comparison. The results of this analysis can be used as a guideline for educational computer game designers in considering appropriate genres of game for educational purpose.

Research related to the first impression has highlighted the importance of visual appeal in influencing the favourable attitude towards a website. In the perspective of impression formation, it is proposed that the users are actually attracted to certain characteristics or aspects of the visual properties of a website, while ignoring the rests. Therefore, this study aims to investigate which visual strongly appeals to the users by comparing the impact of common visuals with the persuasive visuals. The principles of social influence are proposed as the added value to the persuasiveness of the web visuals. An experimental study is conducted and the PLS-SEM method is employed to analyse the obtained data. The result of the exploratory analyses demonstrated that the structural model has better quality when tested with persuasive data sample compared to non-persuasive data sample, evident with stronger coefficient of determination and path coefficients. Thus, it is concluded that persuasive visual provides better impact towards users’ attitude and behavioural intention of a website.

This paper describes an integrated system based on open-domain and domain-specific knowledge for the purpose of providing query-based intelligent web interaction. It is understood that general purpose conversational agents are not able to answer questions on specific domain subject. On the other hand, domain specific systems lack the flexibility to handle common sense questions. To overcome the above limitations, this paper proposed an integrated system comprises of an artificial intelligent conversation software robot or chatterbot, called Artificial Intelligence Natural-language Identity (hereafter, AINI), and an Automated Knowledge Extraction Agent (AKEA) for the acquisition of real world knowledge from the Internet. The objective of AKEA is to retrieve real world knowledge or information from trustworthy websites. AINI is the mechanism used to manage the knowledge and to provide appropriate answer to the user. In this paper, we compare the performance of the proposed system against two popular search engines, two question answering systems and two other conversational systems.

Information and Communications Technologies (ICTs), especially the Internet, have become a key enabler for government organisations, businesses and individuals. With increasing growth in the adoption and use of ICT devices such as smart phones, personal computers and the Internet, Cybersecurity is one of the key concerns facing modern organisations in both developed and developing countries. This paper presents an overview of cybersecurity challenges in Bhutan, within the context that the nation is emerging as an ICT developing country. This study examines the cybersecurity incidents reported both in national media and government reports, identification and analysis of different types of cyber threats, understanding of the characteristics and motives behind cyber-attacks, and their frequency of occurrence since 1999. A discussion on an ongoing research study to investigate cybersecurity management and practices for Bhutan's government organisations is also highlighted.