The ANN is a commonly used network for pattern recognition, and has been trained for various tasks such as prediction, classification, and engineering. However, this model faces challenges such as local minima and slow convergence, which have been addressed through different strategies such as combining the artificial neural network (ANN) with optimised models like the cuckoo search (CS) algorithm. However, for large datasets, the hybrid ANN-based CS algorithm can lead to overfitting. To overcome this issue, the authors propose a new algorithm called Principal Component Analysis with Cuckoo Search Neural Network (PCACSNN). The performance of this algorithm is compared to other commonly used algorithms such as ANN, backpropagation neural network (BPNN), and cuckoo search backpropagation (CSBP), using the Mean Square Error (MSE) and accuracy on classification problems. The simulations were performed on the Student Performance dataset taken from the UCIMLR. The results show that the proposed model performs better than the other models, achieving high accuracy and low MSE for both mathematics and Portuguese student datasets. For the mathematics students, the suggested model attained an accuracy of (99.32%) with MSE of 2.77E-07 for 70% training data and an accuracy of 98.52% with MSE of 2.50E-04 for 30% training data. Similarly, for the Portuguese student dataset, the proposed model obtained (99.38%) accuracy with MSE of 1.09E-08 for 70% training data and 98.72% accuracy with MSE of 1.01E-04 for 30% training data.

Elman Neural Network (ENN) is one of the most common type of recurrent neural network (RNN). It is frequently applied to numerous applications such as classification and predication. However, the ENN model, suffers from serious flaws like network stagnancy and delayed convergence during training. To enhance the effectiveness of the ENN and address the mention problems, numerous studies have been conducted to speed up the learning process of ENN. Many optimization techniques are being used and each algorithm has its own characteristics in terms of efficiency. Some of the algorithms inspire from nature, while others come from the swarm's combined behaviour. Different numbers of optimization methods, including Particle Swarm Optimization (PSO), Bat Algorithm, and Cuckoo Search (CS) are applied to improve the ENN's learning process. As a result, this study suggested using Salp Swarm algorithm (SSA) to address these issues and speed up the ENN algorithm's learning process. The SSA, which is based on the typical behaviour of Salp swarms, is proposed in this research as a novel optimization technique. Due to its effectiveness and Salp intelligent behaviour, a novel Salp algorithm improves the learning process of the artificial neural network (ANN), and ENN models. Therefore this study proposed Salp swarm artificial neural network (SSANN) and Salp swarm Elman neural network (SSElmanNN). Performance of the suggested models is evaluated against ANN, back propagation neural network (BPNN), along with ENN in term of accuracy and Mean Square Error (MSE). Two datasets such as IRIS and Credit Card are used for simulation. The simulation results demonstrate that the suggested models outperform the other algorithms used in this study in terms of MSE and accuracy.

Diabetes has been an offensive condition in recent years, and it can lead to major health problems. If diabetes is not addressed, it can lead to a variety of health problems, including heart disease, stroke, blindness, and kidney failure. Diabetes condition must be addressed promptly to avoid a significant health risk. Machine learning algorithms can assist the doctor in identifying and diagnosing diabetes and other diseases. Different types of classifiers have been used to diagnose diabetes. To improve the performance of integrated flexible individual classifiers and lower the possibility of misclassifying a single instance, an ensemble approach named "Stacking Classifier" was developed. Several classifiers, such as Naïve Bayes, KNN, Linear regression, and decision tree (DT) were used but all these models have low accuracy. However, additional study is needed to detect diabetic condition due to a lack of major work and low accuracy. Therefore this study proposed an ensemble technique termed "Stacking Classifier" was designed to increase the performance of integrated flexible individual classifiers and reduce the probability of misclassifying a single instance. This study uses a variety of classifiers, including Naïve Bayes, KNN, Linear Discriminant Analysis, and Decision Tree, with Random Forest functioning as a Meta classifier. In terms of F-measure, Recall, Accuracy, and Precision, the proposed stacking classifier achieves a higher accuracy of 97.35 % when compared to current models such as Nave Naïve Bayes, KNN, Decision Tree, and Linear Discriminant Analysis, which are 74.60 %, 78.57 %, and 77.35 %, respectively.

Waste management is generally handled by the local municipal authority through traditional approaches, but there exist many limitations of these approaches. Considerable number of research has been performed on solid waste management system. However, smart system especially for streamlet waste (liquid) management is unavailable. In this study, we developed an efficient management system for streamlet using Internet of Things. This system informs the concerned municipal authority in real time about the current waste levels of streamlets in a city. For this purpose, sensor kits are deployed over the streamlets that transmit data to the cloud server for storage, processing and analysing via the Internet. The municipal authority can visualise the data in graph-form entries along with a sensor location from the cloud. When the waste level reaches a predefined threshold value, the waste collector is automatically notified through a smart phone in real time. The performance and accuracy of the proposed system is tested based on the conducted case studies. Experimental results show that the proposed system efficiently delivers its desired operations. The implementation of such system in the society will enhance the quality of life and will make opportunities for government and businessmen to invest money on such a project.

Data classification is one of the most frequently used tasks carried out to label information into predefined classes. The most commonly used models for data classification are Feed Forward Artificial Neural Networks (FFANN), and recurrent neural networks. The connection paths of these two structures are different from each other. Generally, the trained algorithm for these structures is back propagation (BP) algorithm which has many defects. For instance, due to the uncertain number of hidden layer neuron and fixed learning rate, it is easy to fall into local minimum, and it will never reach global minimum error function, it may stay at local minimum. Therefore, to make the slow learning process faster, it is necessary to carefully select the initial weight value. Therefore Meta heuristic search techniques play an important role for selection initial weights for the network. Chicken Swarm Optimization algorithm is one of the Meta heuristic technique effectively for selecting the initial weights values to converge to the optimal solution. However, this study proposed the Chicken Swarm Optimization for efficiently learn the initial weights value of Elman Neural Network algorithm. To validate the proposed algorithm, it is compared with existing algorithms such as Back Propagation Neural Network, Artificial Bee Colony Back Propagation and Genetic Algorithm Neural Network, and verified by two classification problems namely: IRIS and 7-bit parity. Simulation results shows that proposed algorithm outperforms with existing algorithms in terms of accuracy and mean square error.

The pandemic of Covid-19 has caused a shift of paradigm of education, from face-to-face to e-learning. E-learning leads to an escalation in digitalization of handwritten documents because it requires submission of homework and assignments through online. To help teachers in checking digitalized handwritten homework, this paper proposes an automatic checking system based on a convolutional neural network (CNN) for handwritten numeral recognition. The CNN is used to recognize four arithmetic operations in mathematical questions consisting of addition, deduction, multiplication and division. The performance CNN in handwritten numeral recognition have been optimized in terms of activation function and gradient descent algorithm. The proposed CNN is also trained and tested with the MNIST handwritten data set. The experimental results show that the recognition accuracy the improved CNN improves to a certain extent as compared to before optimization.

Stroke is the prominent cause of morbidity and mortality among people worldwide. This study focused to solve the problem of classifying the embolus and artefact from the segmented high intensity transient signal (HITS) by proposing weakly semi-supervised learning approach with active learning. Here, the active learning was executed to select the least confident data for expert labeling. This strategy really improves in detecting embolus from unrestricted topology area and at the same time minimizing the need of a large numbers of label data whilst keeping its performance. Data employed for this study was collected from in-vitro experimental setup. A total of 540 HITS has been used for evaluation. The experimental results proved the advantages of our proposed classification system which showing promising results compared to baseline systems by giving 82.52% and 93% of accuracy and sensitivity, respectively. The system was able to only use a small amount of training data which can reduce the cost of labeling. Furthermore, the proposed system can be adaptively trained using data from different arteries i.e. posterior cerebral artery (PCA), middle cerebral artery (MCA) and internal carotid artery (ICA).

Spiny lobsters are targeted in capture fisheries due the high market demand and this situation has led to overfishing activities. Therefore, a smart monitoring system is imperative to be developed in handling this issue in order to store the distribution of marine spiny lobsters’ information such as habitat preferences, population density and biology aspects of the lobsters. This study focused on separating the mixed underwater acoustic sound. An output of the separation system was used for Passive Acoustic Monitoring (PAM) application. The objective of this study is to provide estimated source signals from a recorded mixed acoustic signal. The feasibility of extracting spiny lobster sound as the target sound was observed during the experiment. In this study, Blind Source Separation (BSS) approach was employed to estimate the target sound (i.e., spiny lobsters) from the mixed underwater acoustic signal consisting of the sound of spiny lobsters and several man-made and natural sounds. In this investigation, two different blind source separation methods namely Fast Fixed Point Independent Component Analysis (FastICA) and Non Negative Matrix Factorization (NMF) were implemented. The mixture of one target sound which was spiny lobster and four interferences signals were used as input in this experiment,. The separated source sound by using FastICA with Negentropy, FastICA with Kurtosis and NMF algorithms were compared and evaluated based on bss_eval_sources toolbox metrics. These metrics consisted of signal to distortion ratio (SDR), signal to interferences ratio (SIR) and signal to artifacts ratio (SAR). In conclusion, the FastICA with Negentropy technique provides the best performance in separating mixed signal based on SIR, SAR and SDR measurement results that showed the FastICA with Negentropy generated the highest average values compared to the FastICA with Kurtosis and NMF.

Technology play an important role in today era. Tremendous growth has been observed in the field of communication and technology. Same is noteworthy in case of video and multimedia. An enormous amount of visual clips containing all types of scenes can easily be accessed both online and offline. In the use of internet has increase the trend of watching cartoons as a source of entertainment among children. There is a tendency that cartoon video has scenes that make some people uncomfortable because of the contents of the video that may be against their feelings and acceptability levels. There could be a number of objectionable elements in cartoon videos. This work presents a framework based on knowledgebase, with visual features which is divided into two categories i.e. character and object database. That deal with the detection of violent contents in cartoon videos which can have negative impact on child personality. A prototype of this system is developed and implemented in Mat lab. The experimental results show that the proposed system successfully detects the violent scenes in cartoon videos.

Increase in global temperature recently has led to heat stroke which can greatly harm the human body. This paper studies the development of a wearable IoT-based heat stroke detection device. This device measured heartbeat rate, surrounding temperature, relative humidity and core body temperature by using several sensors. It generated heat stroke risk level via Fuzzy controller and an alert module alerted the user. The IoT part of this invention has included ThingSpeak server and an android application to store, visualize and display physiological data numerically and graphically. There was 5-phase performance test carried out and the phases of the test were standing, walking, running, walking and standing to evaluate device performance. The results showed that thermal heat stroke risk coefficient (THSRC) was the main factor in the detection of potential heat stroke followed by the other two factors which were core body temperature and heartbeat rate. Based on the results, it is concluded that this device is capable to detect potential heat stroke and alert the user early before any occurrence of heat stroke. It can be inferred that the potential heat stroke will only be detected when the heat stroke risk level is above 25 under the conditions of THSRC of above 37.5, core body temperature of above 38.5 degree Celsius and heartbeat rate of above 135 beats per minute. As an advantage, this wearable IoT-based heat stroke detection device enables user to carry out any activity in hot environment without any worries of being stricken with the heat stroke as the device will immediately alert the user first whenever there is potential heat stroke detected.

Electrocardiogram (ECG) test is very important for fetus condition inspection so as to avoid stillbirth and neonatal death during pregnancy. It is well known and widely used as a medical tool as it is convenient and non-invasive. Nevertheless, analysing fetal ECG (FECG) signal by using naked eye is tedious as the observed signal is a mixed signal which consists of weak FECG, maternal ECG (MECG) and also other signals including mother’s respiratory noise. Hence, in this paper, Blind Source Separation (BSS) is used to extract the estimated desired signal i.e. FECG from the mixed signal. BSS is a well-known separation method that is able to extract desired signal without knowing any information of the source signal. The aim of this study is to elucidate the performance of BSS algorithms i.e. Fast Fixed-Point for Independent Component Analysis (FastICA), Joint Approximate Diagonalization of Eigenmatrix (JADE) and Principal Component Analysis (PCA) for FECG extraction. We integrate R-peak detection algorithm as a post-separation process to the BSS system in order to distinguish the estimated FECG and MECG for easier analysis process. Estimated signals are evaluated based on waveform characteristics observation and Signal-to Interference Ratio (SIR) parameter. We can conclude that JADE performance performs better in term of accuracy while FastICA is good in term of the computational time. However, FastICA manages to get comparable result as JADE after many fine-tuning steps and it is more flexible compared to JADE as it is not sensitive to low quality input signal.

In the past decade, i-vector shows promising result in speaker recognition by modelling a total variability subspace. In this study, as ECG is analogous to speech signals, we focused on the performance of i-vector models based on various compensation methods for ECG identification system in order to combat variability issues. The performance of the models is evaluated based on unbiased protocol (protocol 1) and all-subject protocol (protocol 2) for different compensation methods, including whitening, Linear Discriminate Analysis (LDA), Within-Class Covariance Normalization (WCCN) for single approach, and WCCN-whitening, LDA-whitening, WCCN-LDA for sequential approach. ECG-ID database from Physionet is used for evaluation. It contains 90 subjects with 310 ECG recordings. From the experimental results, we observed that sequential approach has overwhelmed the single approach. i-vector with WCCN-LDA model showed the best rank-1 performance among all the compensation methods for both protocols, 91.89% and 88.89% respectively. In contrast, LDA achieved 67.11% and 64.44% for protocol 1 and 2 respectively, which is the worst among others. We also observed that, the sequential approach requires lesser Gaussian components, consistently 16 components compared to single compensation techniques which requires 32 or more components to achieve the best result hence this shortens the system computation time.

Monitoring the health of fetus at early stage is very crucial. One of the non-invasive ways is by evaluating the pattern of electrocardiogram (ECG) signals of mother’s abdomen and thorax. As these raw signals are mixed signals that consist of mainly maternal ECG (MECG) and fetal ECG (FECG), an effective extraction method of FECG from the mixed signals is imperative. Fast Independent Component Analysis (FastICA) is one of the common signal processing algorithms for blind source separation (BSS). However, it works only for even-determined case when the number of sources (MECG and FECG) is equal to the number of mixtures (two mixed signals i.e. from the mother’s abdomen and the mother’s thorax). For the underdetermined case in which the number of sources would be more than the number of mixture (for the case of twins or triplets’ pregnancy), FastICA algorithm fails as the computation of the inverse mixing matrix of ICA is theoritically impossible. Thus, Degenerate Unmixing Estimation Technique (DUET) algorithm which is based on signal recovery sparsity algorithm is implemented in this research so as to discover its feasibility to solve both even-determined and underdetermined cases. From the experimental results, the DUET performance is promising. Although for even-determined case, FastICA is better than DUET performance, DUET is proved to be valuable in solving the underdetermined case problem, yet the low FECG signal to noise ratio (SNR) value is observed reflecting the high interference.

Finger vein biometric trait has been increasingly used for personal verification or identification in security applications. Normally, the sample image of finger vein is captured under uneven illumination due to geometry variation of finger and it is usually in low contrast condition which will affect the extracted pattern of the vein. In order to have a reliable vein pattern extraction, the contrast and sharpness enhancement of finger image is executed before the pattern extraction procedure. Un-sharp Mask technique has been considered as one of good tools for the sharpness and contrast enhancement of images. However, it suffers with two drawbacks for our intended purpose which are halo effect appears around the finger image and over enhancement of noise to the vein image. Due to this problem, this paper proposes a Modified Un-sharp Mask (MUM) with Log-Gabor filter method to enhance the sharpness and contrast of finger vein image. In this study, the Modified Repeated Line Tracking (MRLT) algorithm is then used to extract the features from the enhanced vein pattern. The experimental results show that the extracted pattern of finger vein contains more details of vein and detection of small details of finger vascular is facilitated after applying the proposed finger vein enhancement method.

Blind Source Separation (BSS) is a task of separating a set of source signals from mixed signal without (or very little information) of both the sources and the mixing process. This paper addresses the problem of BSS in bio-acoustic mixed signals. In a noisy acoustic environment, animal species recognition based on vocalization remains a challenging task. In order to robustly recognize the specific species, the source signals of interest need to be separated from the mixed signals. This separation process is a significant pre-processing step before the recognition process takes place. In this paper, three different source separation methods namely Fast Fixed-Point Independent Component Analysis algorithms (FastICA), Principal Component Analysis (PCA) and Non-Negative Matrix Factorization (NMF) are implemented. In this experiment, the mixtures of frog sound signals are used as input. The quality of separated source signals using FastICA, PCA and NMF algorithms are compared and evaluated according to BSS_EVAL toolbox metrics. These metrics consist of signal to distortion ratio (SDR), signal to interference ratio (SIR) and signal to artifacts ratio (SAR). The results show that FastICA with negentropy technique for finding a maximum non-gaussianity has the best performances in separating mixed signals.

In the domain of speaker recognition, many methods have been proposed over time. The technology for automatic speaker recognition has now reached a good level of performance but there is still need of improvement. In this paper, a new low-dimensional speaker- and channel-dependent space is defined using a simple factor analysis also known as i-vector. This space is named the total variability space because it models both speaker and channel variabilities. The i-vector subspace modelling is one of the recent methods that have become the state of the art technique in this domain. This method largely provides the benefit of modelling both the intra-domain and inter-domain variabilities into the same low dimensional space. In this study, 2656 syllables bio-acoustic signals from 55 species of frog taken from Intelligent Biometric Group, USM database are used for frog identification system. Parameters of the system are initially tuned such as Universal Background Model (UBM) size (32, 64 and 128 Gaussians) and i-vector dimensionality (100, 200 and 400 dimensions). To the end, we assess the effect of the parameter tuned and record the computation time. We observed that, the accuracy for smaller UBM size and higher i-vector dimensionality outperforms others with result of 91.11% is achieved. From this research, it can be concluded that UBM size and i-vector dimensionality effect the accuracy of frog identification based on i-vector.

Ischemic stroke is caused by an occurrence of emboli which travels along the blood vessel in cerebral arteries that eventually trapped near the vessel wall and become stenosis. Transcranial Doppler (TCD) Ultrasound has been used as tool for manual detection of emboli, but the monitoring process is time-consuming and it requires human expert to perform the task. Due to the limited number of experts, this makes manual emboli detection becomes a challenging task. Recently, many researches has devoted to the development of automated emboli detection. In this paper, we investigate the use of frequency and time domain calculations to automatically detect the emboli. In the first method, sinusoidal modelling (SM) is employed to inspect the spectrum of high magnitude frequency component. The second method uses the energy and zero crossing rate (E+ZCR) method. While, the third method is short time energy and short time average zero crossing rate (STE+STAZCR). The experimental results reveal that the sinusoidal modelling gives the best results with genuine acceptance rate is achieved at 84.2%. However, this study also exposes that each approach has its own advantage hence this investigation spurs many rooms of future investigation.

The computation time for solving classification problem using sparse representation classifier remains a huge drawback as it is to be implemented in real time applications. The time consuming of sparse representation classifier is mainly due to the sparse signal recovery solver which is based on l1 minimization or Basis Pursuit. Since then, a fast version of sparse signal recovery solver is introduced and it is based on smoothing the discontinuous properties of l0 norm. In this work, a smoothed l0 norm solver is implemented in sparse representation classifier algorithm. This smoothed l0 norm solver is also modified and improved in such a way to increase its classification accuracy and to further reduce the computation time. The use of kernel version of sparse representation classifier to this modified solver is also implemented and described in this paper. Experiments based on human speech data are carried out in order to compare the improved version of sparse representation classifier with the state of the art classifiers. Experimental results prove that the computation time for classification using proposed algorithm is greatly reduced compared to the baseline performances.

Sparse representation based classification (SRC) has been widely used for pattern recognition especially to face recognition due to its robustness to illumination change, noise and occlusion in face images. SRC method emphasizes the role of parsimonious representation in achieving robustness and accurate classification. To enforce sparsity, the linear representation of query sample and training samples is computed using l1-minimization which is complex and time consuming. Recently, many studies have proved the robustness of SRC is achieved by the collaborative representation mechanism and not the l1 sparsity constraint. Thus, the l1-norm based representation in SRC classification framework could be replaced by l2-norm based representation which is computationally more efficient. This type of classification method is called Collaborative representation based classification (CRC) in the literature. In this paper, an output weight computed from extreme learning machine ELM regarded as a class membership is utilized in conjunction with the classification decision in collaborative representation classifier to improve high classification accuracy over various public available face and speech recognition datasets. The role of ELM is to provide the class membership which denotes the nonlinear similarity of the query sample to training samples from each class. Whereas the CRC provides the linear representation of the query sample and training. The collaborative representation from CRC and class membership from ELM are applied in the regularized residual classification decision to classify the query sample. Experimental results prove that the classification accuracy of the proposed algorithm i.e. CRC-ELM is greatly improved the baseline performances.

Automated security is one of the major concerns in modern time where secure and reliable authentication is in great demand. However, traditional authentication methods such as password and smart card are now outdated because they can be lost, stolen and shared. In this project, biometric system based on heartbeat signals which is also known as Electrocardiographic (ECG) signals is proposed. Heartbeat is chosen as modality due to an individual's ECG signals cannot be faked. Compared to fingerprint it can be fooled with fake fingers, face can be extracted using user's photo and voice can be imitated conveniently. As ECG signals are reflection of the mechanical movement of the heart, these features contain unique physiological information which make them a promising authentication technology. In this study, we develop a portable ECG detection kit for data acquisition. The prototype has successfully tested as a wearable bracelet heartbeat detection for personal system log in. For the software part, wavelet transform algorithm is used as feature extraction technique while for the classification process Support Vector Machine (SVM) is employed. Consequently, the whole system is then integrated on Intel Embedded N2600 Processor with Altera Cyclone IV FPGA Board (DE2-150) as biometric system. Experiment results showed that 2.0069% of EER performance has been achieved, thus this shows that the developed prototype can be a promising modality for biometric system.

Automated identification of animals based their acoustic sound is now preferable by biologist in assisting them to identify animal species for environmental monitoring work. This approach is gradually replacing manual techniques that claimed to be costly and time-consuming. However, it is a challenging task to execute the automated system when the environment is in noisy condition especially in the presence of non-stationary noises such as insect sounds or multiple animal sounds from different species. In this paper, a combination of enhanced start and end point detection namely short time energy (STE) and short time average zero crossing rates (STAZCR) is proposed to improve the syllable segmentation. In this approach, a novel peak finding algorithm is integrated to iteratively narrow down the numbers of local minima and maxima in order to determine the true local maximum value. In this study, the bioacoustics sound samples from frog call database, consists of six hundred and seventy-five frog call data from 15 frog species, recorded in forests located in Kulim and Baling, Malaysia are used. The experimental results demonstrate that 94.13% of performance is achieved by using the proposed method i.e. combination of STE and STAZCR compared to 81.6% of performance for the baseline method, i.e. the combination of the energy and ZCR.