Recent Advances in Information and Communication Technology

Predicting the influence of drinking water pumping on stream and groundwater levels is essential for sustainable water management. Given the highly dynamic nature of such systems any quantitative analysis must be based on robust and reliable modeling and simulation approaches. The paper presents a wireless mesh-network framework for environmental real time monitoring integrated with a cloud computing environment to execute the hydrogeological simulation model. The simulation results can then be used to sustainably control the pumping stations. The use case of the Emmental catchment and pumping location illustrates the feasibility and effectiveness of our approach even in harsh environmental conditions.

Attribute reduction is used to allow elimination of redundant attributes while remaining full meaning of the original dataset. Rough sets have been used as attribute reduction techniques with much success. However, rough set applies to attribute reduction are inadequate at finding optimal reductions. This paper proposes an optimal attribute reduction strategy relying on rough sets and discrete firefly algorithm. To demonstrate the applicability and superiority of the proposed model, comparison between the proposed models with existing well-known methods is also investigated. The experiment results illustrate that performances of the proposed model when compared to other attribute reduction can provide comparative solutions efficiently.

In this paper we present a new clustering algorithm based on a new heuristic we call Chameleon Army. This heuristic simulates a Army stratagem and Chameleon behavior. The proposed algorithm is implemented and tested on well known dataset. The obtained results are compared to those of the algorithms K-means, PSO, and PSO-kmeans. The results show that the proposed algorithm gives better clusters.

The particle swarm optimization (PSO) is an algorithm that attempts to search for better solution in the solution space by attracting particles to converge toward a particle with the best fitness. PSO is typically troubled with the problems of trapping in local optimum and premature convergence. In order to overcome both problems, we propose an improved PSO algorithm that can re-initialize particles dynamically when swarm traps in local optimum. Moreover, the particle re-initialization period can be adjusted to solve the problem appropriately. The proposed technique is tested on benchmark functions and gives more satisfied search results in comparison with PSOs for the benchmark functions.

This paper introduces an

ensemble k-nearest neighbor with neuro-fuzzy

method for the classification. A new paradigm for classification is proposed. The structure of the system includes the use of neural network, fuzzy logic and k-nearest neighbor. The first part is the beginning stages of learning by using 1-hidden layer neural network. In stage 2, the error from the first stage is forwarded to Mandani fuzzy system. The final step is the defuzzification process to create new dataset for classification. This new data is called "

transformed training set

". The parameters of the learning process are applied to the test dataset to create a "

transformed testing set

". Class of the transformed testing set is determined by using k-nearest neighbor. A variety of standard datasets from UCI were tested with our proposed. The fabulous classification results obtained from the experiments can confirm the good performance of

ensemble k-nearest neighbor with neuro-fuzzy

method.

Monthly rainfall spatial interpolation is an important task in hydrological study to comprehensively observe the spatial distribution of the monthly rainfall variable in the study area. A number of spatial interpolation methods have been successfully applied to perform this task. However, those methods mainly aim at achieving satisfactory interpolation accuracy and they disregard the interpolation interpretability. Without interpretability, human analysts will not be able to gain insight of the model of the spatial data. This paper proposes an alternative approach to achieve both accuracy and interpretability of the monthly rainfall spatial interpolation solution. A combination of fuzzy clustering, fuzzy inference system, genetic algorithm and modular technique has been used. The accuracy of the proposed method has been compared to the most commonly-used methods in geographic information systems as well as previously proposed method. The experimental results showed that the proposed model provided satisfactory interpolation accuracy in comparison with other methods. Besides, the interpretability of the proposed model could be achieved in both global and local perspectives. Human analysts may therefore understand the model from the derived model’s parameters and fuzzy rules.

Since 80% of all information in the World Wide Web (WWW) is in textual form, most of the search activities of the users are based on groups of search words forming queries that represent their information needs. The quality of the returned results -usually evaluated using measures such as precision and recall- mostly depends on the quality of the chosen query terms. Therefore, their relatedness must be evaluated accordingly using and matched against the documents to be found. In order to do so properly, in this paper, the notion of n-term co-occurrences will be introduced and distinguished from the related concepts of n-grams and higher-order co-occurrences. Finally, their applicability for search, clustering and data mining processes will be considered.

Variable length time series motif discovery has attracted great interest in the community of time series data mining due to its importance in many applications such as medicine, motion analysis and robotics studies. In this work, a simple yet efficient suffix array based variable length motif discovery is proposed using a symbolic representation of time. As motif discovery is performed in discrete, low-dimensional representation, the number of motifs discovered and their frequencies are partially influenced by the number of symbols used to represent the motifs. We experimented with 4 electrocardiogram data sets from a benchmark repository to investigate the effect of alphabet size on the quantity and the quality of motifs from the time series classification perspective. The finding indicates that our approach can find variable length motifs and the discovered motifs can be used in classification of data where frequent patterns are inherently known to exist.

A similarity of printed Thai characters is a grand challenge of optical character recognition (OCR), especially in case of a variety of font types, sizes, and styles. This paper proposes an effective feature extraction, adaptive histogram of oriented gradient (AHOG), for overcoming the character similarity. The proposed method improves the conventional histogram of oriented gradient (HOG) in two principal phases, which are (i) adaptive partition for gradient images and (ii) adaptive binning for oriented histograms. The former is implemented with quadtree partition based on gradient image variance so as to provide for an effective local feature extraction. The later is implemented with non-uniform mapping technique, so that the AHOG descriptor can be constructed with minimal errors. Based on 59,408 single character images equally divided into training and testing samples, the experimental results show that the AHOG method outperforms the conventional HOG and state-of-the-art methods, including scale space histogram of oriented gradient (SSHOG), pyramid histogram of oriented gradient (PHOG), multilevel histogram of oriented gradient (MHOG), and HOG column encoding algorithm (HOG-Column).

This paper attempted to develop and validate a tool to predict the immediate results of radiation on bone metastasis in cervical cancer cases. Cases of bone metastasis in cervical cancer are based on radiation treatment data, which is imbalanced. This imbalanced data is a challenge among the researchers in data mining, called class imbalance learning (CIL) and has lead to difficulties in machine learning and a reduction in the classifier performance. In this paper, we compared several algorithms to deal with the data imbalance classification problem using the synthetic minority over-sampling technique (SMOTE) used to drive classification models: Ant-Miner, RIPPER, Ridor, PART, ADTree, C4.5, ELM and Weighted ELM using Accuracy, G-mean and F-measure to evaluate performance. The results of this paper show that the RIPPER algorithm outperformed the other algorithms in Accuracy and F-measure, but weighted ELM outperformed other algorithms by G-mean. This may be useful when evaluating clinical assessments.

Drilling operation on bone for screw insertion to fix the broken bones or for the fixation of implants during orthopaedic surgery is highly sensitive. It demands for minimum drilling damage of bone for proper fixation and quick recovery postoperatively. The aim of the present study is to find out an optimum combination of bone drilling parameters (feed rate and spindle speed) for minimum thrust force during bone drilling using genetic algorithm (GA). Central composite design is employed to carry out the bone drilling experiments and based on the experimental results, a response surface model was developed. This model is used as a fitness function for genetic algorithm (GA). The investigation showed that the GA technique can efficaciously estimate the optimal setting of bone drilling parameters for minimum thrust force value. The suggested approach can be very useful for orthopaedic surgeons to perform minimally invasive drilling of bone.

Nowadays, commercial activity on Internet and media requires a protection by increasing security. The 2D Barcode with a digital watermark is a widely interest research in security field. QR Code with invisible watermark prevents information hiding text. This paper proposes QR Code (Quick Response Code) that is embedded an invisible video watermark by using Discrete Multiwavelet transformation (DMT). We have developed an optimization technique using the genetic algorithm to search for optimal quantization step in order to improve both quality of watermarked video and robustness of the watermark. This technique does not require the original image in the watermark extraction. The experimental results show that the proposed watermarking algorithm yields watermarked image with good imperceptibility and very robust.

Mining multi-level frequent itemsets from transactional database is one of the most important tasks in data mining community. It aims to discover correlation among items with their hierarchical categories under support-confidence values and thresholds. However, it is well-known that the task of providing an appropriate support threshold to mine the most interesting patterns without prior knowledge in advance is very difficult and it is more reasonable to ask the users to specify the number of desired patterns. Therefore, in this paper, we propose an alternative approach to mine the most interesting multi-level frequent patterns without the setting of support threshold, called

N-most interesting multi-level frequent pattern mining

, where

N

is the number of desired patterns with the highest support values per each category level. To mine such patterns, an efficient adaptive

FP-growth

algorithm, called

NMLFP

, is proposed. Extensive performance studies show that

NMLFP

has high performance and linearly scalable on the number of desired results.

Content-based image retrieval is an active research area in image processing and computer vision. Color represents an important feature in CBIR applications, thus many color descriptors were proposed. Sequential search is one of the common drawbacks of most color descriptors especially in large databases. In this paper, dominant colors of an image are indexed to avoid sequential search in the database. Dominant colors in query image are used independently to find images that containing similar colors to create reduced search space instead of the whole database search space. This will speed up the retrieval process in addition to improve the accuracy of color descriptors. Experimental results show effectiveness of the proposed color indexing method in reducing the search space to less than 25% without degradation the accuracy.

Recent research suggests that the “big five personality traits” influence the purchasing and usage preferences of mobile application. However, the impact of monetizing of applications and personality traits has so far been largely unattended. We have therefore extended our research to cover monetizing models of mobile applications. In this paper, we aim to enhance the understanding of the relationship between the “big five personality traits” and the usages and purchase intention of mobile applications in difference categories. Our initial data for the pilot study consists of 173 individuals, collected from smart device consumers who live in Bangkok, Thailand. Pearson’s correlation and multiple linear regressions were used to analyze the data. The initial results indicate that some personality traits are associated with the usages and intention to purchase mobile applications. It is highly possible to conclude from the data that conscientious persons placed more intention to use productive applications. Specifically, this personality trait has a positive relationship with utilities, education, business and maps and navigation. Neuroticism reported only significant relation with in-app purchase in utilities applications. Agreeableness showed no significance during our regressions analysis. The most widely used paid application among all traits is entertainment. The findings of this pilot study will serve as indicators for the direction of our planned future research in this field.

This paper proposes an intelligent framework to accurately analyze these echo images in order to discover disease category and assess the severity automatically. Typically, each video consists of 90-100 frames of 2D echo, color Doppler image video, and several .jpg images. Our framework consists of parallel algorithms developed under OpenMP environment. The major tasks are cardiac boundary tracing, quantifying the heart chambers and extracting 2D features, and other features required for computing statistical features and build a classifier model for categorization. Segmentation of an echo image is done using parallel implementation of K-Means algorithm and they are boundary extracted using active contour method. Bayesian model is used to classify a given patient into normal or abnormal. The experiment involves videos taken from 60 normal and abnormal patients from a local cardiology Hospital. The results obtained with our algorithms outperformed with respect to the results that have already been reported.

Durians are green spiky fruits, which are considered as a delicacy throughout Southeast Asia. They are valued for their unique flavor and powerful taste. It is desirable to be able to determine the quality of durians without cutting them because it is difficult to quantify the ripeness from the external appearance and they are expensive to purchase. In Southeast Asia and China, consumers have found that after buying and cutting durians, they are not ripe or ready to eat. Therefore, studying striking signal characteristics and developing an automated method of recognizing durian ripeness levels without cutting or destroying them could benefit consumers of the fruit. The following method of recognizing durian ripeness by studying striking signals using N-gram models with N-best lists and majority voting is proposed. The recognition process is composed of three stages: 1) extract the acoustic features from the striking signals, 2) recognize unripe and ripe durian striking signals using the N-gram models and 3) find the ripeness from the N-best lists using majority voting. The results indicate that using the 3-best lists and majority voting method it was possible to recognize durian ripeness efficiently. Average ripeness recognition rates of 95.8%, 90.4% and 93.1% were obtained from the untrained, unknown and both test sets, respectively. The results demonstrate that the method is accurate enough to be used by consumers to help them select a ripe durian.

Software companies are replacing traditional software development methods with Agile methods due to coping with inherent problems of traditional methods. Due to the different nature of traditional and Agile methods, adaptation to Agile methods is not a simple process and needs to be managed in a sustainable way. In recent years, several studies have conducted on investigation of Agile migration journey, but less effort on identifying the serious managerial attentions in Agile transition process. Conducting a Grounded Theory in context of Agile software development, showed various aspects of the transition to be considered in order to having a successful change management process. This paper shows the important role of the emergent managerial attentions on success of Agile transition and adoption process.

Super-peer topologies have been found efficient and effective in heterogeneous peer-to-peer networks. Due to dominant position of super-peers, super-peer selection requires a protocol that is aware of peer capacities. Lack of global information about other peers’ capacity and dynamic nature of peer-to-peer networks are two major challenges that impose uncertainty in decision-making. SG-1, is a well-known super-peer selection protocol considering peer capacities, but lack of an appropriate decision-making mechanism makes this protocol slow at convergence and imposes overhead of client transfer between selected super-peers. In this paper, we propose an improved version of SG-1 that uses learning automata as an adaptive decision-making mechanism. For this purpose, each peer is equipped with a learning automaton which is used locally in the decisions taken by that peer. Simulations show effectiveness of proposed protocol in terms of convergence time, scalability, capacity utilization, behavior towards super-peer failure and communication cost, compared to SG-1.

In this paper, we present a derivative of

M

idimew connected

M

esh

N

etwork (

MMN

) by reassigning the free links for higher level interconnection for the optimum performance of the MMN; called Derived MMN (DMMN). We present the architecture of DMMN, addressing of nodes, routing of message and evaluate the static network performance. It is shown that the proposed DMMN possesses several attractive features, including constant degree, small diameter, low cost, small average distance, moderate bisection width, and same fault tolerant performance than that of other conventional and hierarchical interconnection networks. With the same node degree, arc connectivity, bisection width, and wiring complexity, the average distance of the DMMN is lower than that of other networks.

Malaysia is considered the eighth Asian country out of the top 15 countries in household broadband penetration at 34.5%. Users in rural areas who cannot receive Digital Subscriber Line (DSL) or cable modem services. In addition, owing to the high cost of Information and Communication Technology (ICT) infrastructure deployment in the rural areas, delay in broadband services is being experienced. Therefore, the Power Lines Communication (PLC) technology could have the potential to provide a broadband access through the entire electricity grid. Broadband PLC uses power lines as a high-speed digital transmission channel. This paper investigates low voltage Channel Transfer Function (CTF) of PLC technology for the purpose of data transmission regarding Malaysia electrical cable specification by using Matlab-Simulink simulation tool. Since that could help Malaysia licensable activities to provide Broadband PLC service out of harm’s way of other data communication networks.

Recently, recommendation systems have become an important tool to support and improve decision making for educational purposes. However, developing recommendation systems is far from trivial and there are specific issues associated with individual problems. Low-correlated input features is a problem that influences the overall accuracy of decision tree models. Weak relationship between input features can cause decision trees work inefficiently. This paper reports the use of features grouping method to improve the classification accuracy of decision trees. Such method groups related input features together based on their ontologies. The new inherited features are then used instead as new features to the decision trees. The proposed method was tested with five decision tree models. The dataset used in this study were collected from schools in Nakhonratchasima province, Thailand. The experimental results indicated that the proposed method can improve the classification accuracy of all decision tree models. Furthermore, such method can significantly decrease the computational time in the training period.

Resources abstractions have been very critical in securing performance improvement and higher acceptance by the end users in cloud computing. System virtualization, storage virtualization and network virtualization had been realized and has become as a part large systems abstraction in cloud computing. Memory high-availability in network environments is an added advantage in our presentation as an integral part in extending cloud computing multi various services to include network memory virtualization.

In this paper, we describe the virtualization of memory in cluster environment that could be applied universally using RDMA utility to map and access memory across the network. We suggested a combination of using the latency of remote memory and direct remote memory mapping facilities in our implementation. A low-level remote memory allocation and replacement technique is introduced to minimize page faulting and provide option to be more fault tolerance. We proposed a low level memory management technique in a network environment which would be able to support Service Oriented Architecture (SOA) and cloud computing.

Follower relations are the new currency in the social web. User-generated content plays an important role for the tie formation process. We report an approach to predict the follower counts of Twitter users by looking at a small amount of their tweets. We also found a pattern of textual features that demonstrates the correlation between Twitter specific communication and the number of followers. Our study is a step forward in understanding relations between social behavior and language in online social networks.

Search engines usually deliver a large amount results for each topic addressed by a few (mostly 2 or 3) keywords. Thus, it is a tough work to find those terms describing the wanted content in a manner such that the search delivers the intended results already on the first result pages. In the iterative process of obtaining the desired web pages, pictures with their tremendous context information may be a big help. This contribution presents an approach to include picture processing by humans as a means for context search selection and determination in a locally working search control.

In multi-class text classification, when number of entities in each class is highly imbalanced, performance of feature ranking methods is usually low because the larger class has much dominant influence to the classifier and the smaller one seems to be ignored. This research attempts to solve this problem by separating the larger classes into several smaller subclasses according to their proximities, by k-mean clustering then all subclasses are considered for feature scoring measure instead of the main classes. This cluster-based feature scoring method is proposed to reduce the influence of skewed class distributions. Compared to performance of feature sets selected from main classes and ground-truth subclasses, the experimental results show that performance of a feature set selected by the proposed method achieves significant improvement on classifying imbalanced corpora, the RCV1v2 dataset.

The realization of semantic web depended on the availability of web of data associated with knowledge and information in the real world. The first stage for web of data preparation is semantic annotation. However framing such manual semantic annotation is inappropriate for inexperienced users because they require specialist knowledge of ontology and syntax. To address this problem, this paper proposes an approach of automatic semantic annotation based on the integration between natural language processing techniques and semantic web technology. A case study on leave of absence correspondence in Thai language is chosen as the domain of interest. Our study shows that the proposed approach verified the effectiveness of semantic annotation.

Traditional Information Retrieval systems (keyword-based search) suffer several problems. For instance, synonyms or hyponym are not taken into consideration when retrieving documents that are important for a user’s query. This study adopts an ontology of computer science and proposes an ontology indexing weight based on Wu and Palmer’s edge counting measure for solving this problem. This paper used the N-grams method for computing a family of word similarity. The study also compares the subsumption weight between Hliaoutakis and Nicola’s weight and query keywords (Decision Making, Genetic Algorithm, Machine Learning, Heuristic). A probability value (p-values) from the t-test (p = 0.105) is higher 0.05 and indicates no significant evidence, of not differences between both weights methods. The experimental results show that the document similarity score between a user’s query and the paper suggests that the new measures were effectively ranked.

One of the most problems facing learners in e-learning system is to find the most suitable course materials or learning objects for their personalized learning space. The main focus of this paper is to extend our previous rule-based representation recommendation system [1] by applying an ontology-based approach for creating a semantic learning object recommendation named ”LORecommendNet”. The ”LORecommendNet” ontology represents the knowledge about learning objects, learner model, semantic mapping rules and their relationship are proposed. In the proposed framework, we demonstrated how the ”LORecommendNet” can be used to enable machines to interpret and process learning object in recommendation system. We also explain how ontological representations play a role in mapping learner to personalized learning object. The structure of “LORecommendNet” extends the semantic web technology, which the representation of each based on an OWL ontology and then on the inference layer, based on SWRL language, making a clarify separation of the program components and connected explicit modules.

The security risks of cloud computing include loss of control over data and programs stored in the cloud, spying out these data and unnoticed changing of user software by the cloud provider, malware intrusion into the server, eavesdropping during data transmission as well as sabotage by attackers able to fake authorised users. It will be shown here how these security risks can effectively be coped with. Only for preventing the cloud provider from wrong-doing no technical solution is available. The intrusion of malware into cloud servers and its malicious effects can be rendered impossible by hardware-supported architectural features. Eavesdropping and gaining unauthorised access to clouds can be prevented by information-theoretically secure data encryption with one-time keys. A cryptosystem is presented, which does not only work with one-time keys, but allows any plaintext to be encrypted by a randomly selected element out of a large set of possible ciphertexts. By obliterating the boundaries between data items encrypted together, this system removes another toehold for cryptanalysis.

Web logs play the crucial role in detecting web attack. However, analyzing web logs become a challenge due to the huge log volume issue. The objective of this research is to create a web log cleaning algorithm for web intrusion detection. Studies on previous works showed that there are five major web log attributes needed in web log cleaning algorithm for intrusion detection, namely multimedia files, web robots request, HTTP status code, HTTP method and other files. The enhanced algorithm is based on these five major web log attributes along with a set of rules and conditions. Our experiment shows that the proposed algorithm is able to clean noisy data effectively with a percentage of reduction of 40.41 and at the same time maintain the readiness for web intrusion detection at a low false negative rate (0.00531). Future works may address the web intrusion detection mechanism.

The aim of this research is to propose a new modified integer factorization algorithm, called Possible Prime Modified Fermat Factorization (P

2

MFF), for breaking RSA which the security is based upon integer factorization. P

2

MFF is improved from Modified Fermat Factorization (MFF) and Modified Fermat Factorization Version 2 (MFFV2). The key concept of this algorithm is to reduce iterations of computation. The value of larger number in P

2

MFF is increased more than one in each iteration of the computation, it is usually increased by only one in MFF and MFFV2. Moreover, this method can decrease the number of times in order to compute the square root of some integers whenever we can strongly confirm that square root of these integers is not an integer by using number theory. The experimental results show that P

2

MFF can factor the modulus faster than MFF and MFFV2.

Typically, video copy detection can be done by comparing signatures of new content with of known contents in database. However, this method requires high computation for both database generation and signature detection. In this paper, we proposed an efficient and fast video signature for video copy protection. The video features of a scene are extracted and then transformed to be a signature as a bit-wise string. All string signatures then are stored and manipulated by n-gram based text retrieval algorithm, which is proposed as a replacement with computation-intensive content similarity detection algorithm. The evaluation on the CC_WEB_VIDEO dataset shows that its accuracy is 85% where our baseline algorithms achieved only 75%; however, our algorithm is around 20 times as fast as the baseline.