Combining Mel Frequency Cepstral Coefficient with wavelet transform for feature extraction is not new. This paper proposes a new architecture to help in increasing the accuracy of speaker recognition compared with conventional architecture. In conventional speaker model, the voice will undergo noise elimination first before feature extraction. The proposed architecture however, will extract the features and eliminate noise simultaneously. The MFCC is used to extract the voice features while wavelet de-noising technique is used to eliminate the noise contained in the speech signals. Thus, the new architecture achieves two outcomes in one single process: ex-tracting voice feature and elimination of noise.
Mel frequency cepstral coefficient, Speaker recognition, Wavelet transform
In recent years, many classification models have been developed and applied to increase their accuracy. The concept of distance between two samples or two variables is a fundamental concept in multivariate analysis. This paper proposed a tool that used different similarity distance approaches with ranking method based on Mean Average Precision (MAP). In this study, several similarity distance methods were used, such as Euclidean, Manhattan, Chebyshev, Sorenson and Cosine. The most suitable distance measure was based on the smallest value of distance between the samples. However, the real solution showed that the results were not accurate as and thus, MAP was considered the best approach to overcome current limitations.
Accuracy, Mean average precision, Ranking, Similarity distance
Space weather forecasting and its importance for the power and communication industry have inspired research related to TEC forecasting lately. Research has attempted to establish an empirical model approach for TEC prediction. In this paper, artificial neural networks (ANNs) have been applied in total electron content using GPS Ionospheric Scintillation and TEC Monitor (GISTM) data from UKM Station. The TEC prediction will be useful in improving the quality of current GNSS applications, such as in automobiles, road mapping, location-based advertising, personal navigation or logistics. Hence, a neural network model was designed with relevant features and customised parameters. Various types of input data and data representations from the ionospheric activity were used for the chosen network structure, which was a three-layer perceptron trained by feed forward back propagation method and tested on the chosen test data. We found that the optimum RMSE occurred with 10 nodes as the best NN for GISTM UKM station for the studied period with RMSE 1.3457 TECU. An analysis was made to compare the TEC from the measured TEC with neural network prediction and from IRI-corr model. The results showed that the NN model forecast the TEC values close to the measured TEC values with 9.96% of relative error. Thus, the forecasting of total electron content has the potential to be implemented successfully with larger data set from multi-centred environment.
Forecasting, GPS, IRI-corr model, Neural network, Total electron content
This paper investigates the accuracy of Feedforward Neural Network (FFNN) with different external parameters in predicting the closing price of a particular stock. Specifically, the feedforward neural network was trained using Levenberg-Marquardt backpropagation algorithm to forecast the CIMB stocks closing price in the Kuala Lumpur Stock exchange (KLSE). The results indicate that the use of external parameters can improve the accuracy of the stocks closing price.
The Gleason grading system assists in evaluating the prognosis of men with prostate cancer. Cancers with a higher score are more aggressive and have a worse prognosis. The pathologists observe the tissue components (e.g. lumen, nuclei) of the histopathological image to grade it. The differentiation between Grade 3 and Grade 4 is the most challenging, and receives the most consideration from scholars. However, since the grading is subjective and time-consuming, a reliable computer-aided prostate cancer diagnosing techniques are in high demand. This study proposed an ensemble computer-added system (CAD) consisting of two single classifiers: a) a specialist, trained specifically for texture features of the lumen and the other for nuclei tissue component; b) a fusion method to aggregate the decision of the single classifiers. Experimental results show promising results that the proposed ensemble system (area under the ROC curve (Az) of 88.9% for Grade 3 versus Grad 4 classification task) impressively outperforms the single classifier of nuclei (Az=87.7) and lumen (Az=86.6).
The false positive (FP) is an over-segment result where the noncancerous pixel is segmented as a cancer pixel. The FP rate is considered a challenge in localising masses in mammogram images. Hence, in this article, a rejection model is proposed by using a supervised learning method in mass classification such as support vector machine (SVM). The goal of the rejection model which is based on SVM is the reduction of FP rate in segmenting mammogram through the Chan-Vese method, which is initialised by the marker controller watershed (MCWS) algorithm. The MCWS algorithm is utilised for segmentation of a mammogram image. The segmentation is subsequently refined through the Chan-Vese method, followed by the development of the proposed SVM rejection model with different window size as well as its application in eliminating incorrect segmented nodules. The dataset comprised of 57 nodules and 113 non-nodules and the study successfully proved the effectiveness of the SVM rejection model to decrease the FP rate.
Breast cancer, Chan-Vese, Mammogram, MCWS, Rejection model, SVM
In this paper, an image binarization method for separating text from the background of degraded textual images is proposed. This proposed methods are based on combination of Window Tracking Method (WTM) and Dynamic Image Histogram (DIH). The WTM and DIH methods work on an image that has been pre-processed. The WTM method searches for the largest pixel value in a 3 Ã— 3 window up to a maximum of five tracking steps, while the method searches for a definite frequency between the two highest values in the image histogram. We test proposed method on DIBCO dataset and self-collection faded manuscripts. The experimental results show that our proposed method outperforms state of the art methods.
The watermarking is a method of concealing digital information in multimedia data, namely the host image. Discrete wavelet transform (DWT) when joined with discrete cosine transform (DCT) and SVD deliver powerful digital watermarking image. There are different types of intrusions that either plunder the actual ownership or demolish the appearance. In this paper, the DWT-DCT, DWT-SVD approach has been proposed to ensure security by concealing the watermark inside the actual image and validate the proprietor's image. Using DWT-DCT and low-bit percentage, the watermark image was inserted and abstracted. The DWT-SVD hybrid produced very good results.
Digital image, Discrete cosine transform (DCT), Discrete wavelet transform (DWT), Singular value decommission (SVD), Watermarking
Police patrol routing problem (PPRP) attracts researchers' attention especially on artifitial inteligence. The challenge here is that a limited number of patrols cover a wide range of area that includes several hotspots. In this study, a new model for PPRP is proposed simulating the Solomon's benchmark for vehicle routing problem with time windows. This model can solve this problem by maximising the coverage of hotspots with frequencies of high priority locations while ensuring the feasibility of routes. Two constructive greedy heuristics are developed to generate the initial solution of the PPRP: highest priority greedy heuristic (HPGH) and nearest neighbour greedy heuristic (NNGH). Experimental results show that the simulated Solomon's benchmark is suitable to represent PPRP. In addition, results illustrate that NNGH is more efficient to construct feasible solution than HPGH.
Simulating Lotka-Volterra model using a numerical method requires the researcher to apply tiny mesh sizes to come up with an accurate solution. This approach will increase the complexity and burden of computer memory and consume long computational time. To overcome these issues, a new solver is used that could simulate Lotka-Volterra model using bigger mesh size. In this paper, prey and predator behaviour is simulated via Lotka-Volterra model. We approximate the nonlinear terms in the model via weighted average approach and differential equation via nonstandard denominators. We provide three new schemes for one step method and simulate four sets of parameters to examine the performance of these new schemes. Results show that these new schemes simulate better for large mesh sizes.
Lotka-Volterra, Nonlinear, Nonstandard method, Weighted average approach
Feature descriptor for image retrieval has emerged as an important part of computer vision and image analysis application. In the last decades, researchers have used algorithms to generate effective, efficient and steady methods in image processing, particularly shape representation, matching and leaf retrieval. Existing leaf retrieval methods are insufficient to achieve an adequate retrieval rate due to the inherent difficulties related to available shape descriptors of different leaf images. Shape analysis and comparison for plant leaf retrieval are investigated in this study. Different image features may result in different significance interpretation of images, even though they come from almost similarly shaped of images. A new image transform, known as harmonic mean projection transform (HMPT), is proposed in this study as a feature descriptor method to extract leaf features. By using harmonic mean function, the signal carries information of greater importance is considered in signal acquisition. The selected image is extracted from the whole region where all the pixels are considered to get a set of features. Results indicate better classification rates when compared with other classification methods.
Simulating Lotka-Volterra model using numerical method require researchers to apply tiny mesh sizes to obtain an accurate result. This approach nevertheless increases the complexity and burden of computer memory and consume long computational time. To overcome these issues, we investigate and construct new two-step solver that could simulate Lotka-Volterra model using bigger mesh size. This paper proposes three new two-step schemes to simulate Lotka-Volterra model. A non-standard approximation scheme with trimean approach was adopted. The nonlinear terms in the model is approximated via trimean approach and differential equation via non-standard denominators. Four sets of parameters were examined to analyse the performance of these new schemes. Results show that these new schemes provide better simulation for large mesh size.
Sentiment analysis classification has been typically performed by combining features that represent the dataset at hand. Existing works have employed various features individually such as the syntactical, lexical and machine learning, and some have hybridized to reach optimistic results. Since the debate on the best combination is still unresolved this paper addresses the empirical investigation of the combination of features for product review classification. Results indicate the Support Vector Machine classification model combined with any of the observed lexicon namely MPQA, BingLiu and General Inquirer and either the unigram or inte-gration of unigram and bigram features is the top performer.
Product review, sentiment classification, sentiment features
There has been an increase of content related to Quran and Hadith on the internet over the past few years. Diacritical Digital Quran is very sensitive to tampering. Diacritics are the symbols used beneath/above Quranic verses for reading purposes of the Quran. Minor change in diacritics can alter the meaning of a particular Quranic verse. Hence, there is a need for an authentication system to differentiate between fake and original verses. In this work, a model is proposed related to automatic authentication of Digital Quran. Authentication model is divided into two phases: tokenisation and authentication. For tokenisation, regular expressions are used to split input Quranic verse into single characters. In case of authentication, existing and standard exact matching algorithm i.e. Quick search (QS) is used. On testing the proposed model by comparing popular search engines and other related existing works, our approach is 100 % accurate in terms of full verse detection.
Diacritical verse, Exact matching, Hadith authentication, Quran authentication and integrity, Quranic verse and text authentication
Over the last few years, the Android smartphone had faced attacks from malware and malware variants, as there is no effective commercial Android security framework in the market. Thus, using machine learning algorithms to detect Android malware applications that can fit with the smartphone resources limitations became popular. This paper used state of the art Deep Belief Network in Android malware detection. The Lasso is one of the best interpretable â„“1-regularisation techniques which proved to be an efficient feature selection embedded in learning algorithm. The selected features subset of Restricted Boltzmann Machines tuned by Harmony Search feature reduction with Deep Belief Network classifier was used, achieving 85.22% Android malware detection accuracy.
One of the major problems in today's economy is the phenomenon of tax evasion. The linear regression method is a solution to find a formula to investigate the effect of each variable in the final tax evasion rate. Since the tax evasion data in this study has a great degree of uncertainty and the relationship between variables is nonlinear, Bayesian method is used to address the uncertainty along with 6 nonlinear basis functions to tackle the nonlinearity problem. Furthermore, variational method is applied on Bayesian linear regression in tax evasion data to approximate the model evidence in Bayesian method. The dataset is collected from tax evasion in Malaysia in period from 1963 to 2013 with 8 input variables. Results from variational method are compared with Maximum Likelihood Estimation technique on Bayeisan linear regression and variational method provides more accurate prediction. This study suggests that, in order to reduce the tax evasion, Malaysian government should decrease direct tax and taxpayer income and increase indirect tax and government regulation variables by 5% in the small amount of changes (10%-30%) and reduce direct tax and income on taxpayer and increment indirect tax and government regulation variables by 90% in the large amount of changes (70%-90%) with respect to the current situation to reduce the final tax evasion rate.
Face detection and analysis is an important area in computer vision. Furthermore, face detection has been an active research field in the recent years following the advancement in digital image processing. The visualisation of visual entities or sub-pattern composition may become complex to visualise due to the high frequency of noise and light effect during examination. This study focuses on evaluating the ability of Haar classifier in detecting faces from three paired Min-Max values used on histogram stretching. Min-Max histogram stretching was the selected method for implementation given that it appears to be the appropriate technique from the observation carried out. Experimental results show that, 60-240 Min-Max values, Haar classifier can accurately detect faces compared to the two values.
Resident's vehicles in some institutions have to be registered to maintain traffic safety. Penalties should be imposed if residents break traffic rules. Most of the time, the vehicle owner's information is difficult to access making the penalty registration process complicated. An effective penalty registration process is required to make the process easier for security officers to give notice to the residents who have committed traffic offenses. A mobile application is proposed to recognize vehicle owner information. The proposed application uses optical-character-recognition (OCR) technologies that can facilitate the process of recognizing vehicle's registration number in order to obtain owner information and use the information to enrol the penalty. The proposed application recognizes the vehicle registration number or sticker serial number to access the owner information. For evaluation of the proposed application, a user study was conducted by asking the users to use the application and answer the questiuonnaire. The findings revealed that average score of 77 of the respondents agree in terms of satisfaction and adoption of the application to be utilized in some institutions. The proposed application reduces the paper work of security officers and makes them more efficient.
Database management system, Number plate recognition, Optical Character Recognition
The walking of a humanoid robot needs to be robust enough in order to maintain balance in a dynamic environment especially on uneven terrain. A walking model based on multi-sensor is proposed for a Robotis DARwIn-OP robot named as Leman. Two force sensitive resistor (FSRs) on both feet equipped to Leman to estimate the zero moment point (ZMP) alongside with accelerometer and gyrosensor embedded in the body for body state estimation. The results show that the FSRs can successfully detect the unbalanced walking event if the protuberance exists on the floor surface and the accelerometer and gyrosensor (Inertial Measurement Unit, IMU) data are recorded to tune the balancing parameter in the model.
This paper deals with the analysis of different Fuzzy membership type performance for Extended Kalman Filter (EKF) based mobile robot navigation. EKF is known to be incompetent in non-Gaussian noise condition and therefore the technique alone is not sufficient to provide solution. Motivated by this shortcoming, a Fuzzy based EKF is proposed in this paper. Three membership types are considered which includes the triangular, trapezoidal and Gaussian membership types to determine the best estimation results for mobile robot and landmarks locations. Minimal rule design and configuration are also other aspects being considered for analysis purposes. The simulation results suggest that the Gaussian memberships surpassed other membership type in providing the best solution in mobile robot navigation.
Fuzzy logic, Kalman Filter, Membership, Mobile robot, Navigation
The water flow-like algorithm (WFA) is a relatively new metaheuristic algorithm, which has shown good solution for the Travelling Salesman Problem (TSP) and is comparable to state of the art results. The basic WFA for TSP uses a 2-opt searching method to decide a water flow splitting decision. Previous algorithms, such as the Ant Colony System for the TSP, has shown that using k-opt (k>2) improves the solution, but increases its complexity exponentially. Therefore, this paper aims to present the performance of the WFA-TSP using 3-opt and 4-opt, respectively, compare them with the basic WFA-TSP using 2-opt and the state of the art algorithms. The algorithms are evaluated using 16 benchmarks TSP datasets. The experimental results show that the proposed WFA-TSP-4opt outperforms in solution quality compare with others, due to its capacity of more exploration and less convergence.
Combinatorial optimization, Nature-inspired metaheuristics, Traveling Salesman Problem, Water flow-liked algorithm
K-Means is an unsupervised method partitions the input space into clusters. K-Means algorithm has a weakness of detecting outliers, which have it available in many variations research fields. A decade ago, Rough Sets Theory (RST) has been used to solve the problem of clustering partition. Specifically, Rough K-Means (RKM) is a one of the powerful hybrid algorithm, which has it, has various extension versions. However, with respect of the ideas of existing rough clustering algorithms, a suitable method to detect outliers is much needed now. In this paper, we propose an effective method to detect local outliers in rough clustering. The Local Outlier Factor (LOF) method in rough clustering improves the quality of the cluster partition. The improved algorithm increased the level of clusters quality. An existing algorithm version, the Ï€ Rough K-Means (Ï€ RKM) tested in the study. Finally, the effectiveness of the algorithm performance is demonstrated based on synthetic and real datasets.
Data analysis, Local outlier factor, K-Means; Rough clustering, Outlier detection
Blood cancer is an umbrella term for cancers that affect the blood, bone marrow and lymphatic system. There are three main groups of blood cancer: leukemia, lymphoma and myeloma. Some types are more common than others. In this paper, a new image transform based on geometric mean properties of integral values in both horizontal and vertical image directions is proposed for leukemia cancer cell classification. Available classification methods using the classical feature extraction methods which are sensitive to rotation and deformation of the blood cells. The new transform is based on geometric mean projection, which â€”unlike other image transforms, such as Radon transformâ€” is not considered all signals in an image with the same signal acquisition rate. Instead, it is general and thus applicable to all capturing signal functions to achieve sufficient invariant features. The geometric mean projection transforms (GMPT) guarantees that the detector only extracts the highly informative information from the object to achieve an invariant feature vector for an accurate classification process. This method has been used as cancer cell identification using microscopic Imagery analysis in this study. Dissimilarity metric calculation and shape analysis by using image transform has been used to extract the feature vectors of the imagery. Then, the accumulated feature vectors have been classified to different classes by using artificial neural network (ANN). The proposed technique has been evaluated in the standard images sourced from USIM, Malaysia. The evaluation results indicate the robustness of the technique in different types of images available in the dataset.
Cancer cell classification Image transform, Image processing, Pattern recognition
Visual Simultaneous Localization and Mapping (vSLAM) system is widely used by autonomous mobile robots. Most vSLAM systems use cameras to analyze surrounding environment and to build maps for autonomous navigation. For a robot to perform intelligent tasks, the built map should be accurate. Landmark features are crucial elements for mapping and path planning. In the vSLAM literature, loop closure detection is a very important process for enhancing the robustness of the vSLAM algorithms. The most widely used algorithms for loop closure detection use a single descriptor. However, the performance of the single descriptors appears to worsen as the map keeps growing. One possible solution to this problem is to use multiple descriptors and combine them as in Naive and linear combinations. These approaches, however, have weaknesses in recognizing the correct locations due to overï¬tting and high-bias, which hinder the generalization performance. This paper proposes the usage of ensemble learning to combine the predictions of multiple Bayesian ï¬lter models which make more accurate prediction than individual models. The proposed approach is validated on three public datasets, namely, Lip6 Indoor, Lip6 Outdoor and City Centre. The results show that the proposed ensemble algorithm signiï¬cantly outperforms the single approaches with a recall of 80%, 97% and 87%, with 100% precision on the three datasets, and outperforms the Naive approach and the existing loop closure detection algorithms.