International Journal of Intelligent Engineering and Systems

1

Authors: Sreerama Prasad Chelluboina, Kunjum Nageswara Rao

Tongue diagnosis is an important way of diagnosing the human health in Indian ayurvedic medicine (IAM), which helps to identify different diseases through tongue image analysis. Traditionally, this diagnosis system has been done by ayurvedic experts to recognize the type of health issue by observing the human tongue. But the recent advancements in machine learning algorithms in medical field enabled the researchers to implement a computer aided diagnosis system for ayurvedic medical treatment. However, the existing tongue diagnosis systems are suffering with poor classification performance due to the variations in tongue appearance such as color, shape, coating, and texture properties. Therefore, this article focuses on computer aided tongue diagnosis system (CATDS) for disease predication through tongue image analysis, here after denoted as CATDSNet. Initially, fast nonlocal mean (FNLM) filtering is applied on test image to perform the preprocessing of given tongue dataset. Next, color features are extracted from the pre-processed tongue images using colour moments. In addition, grey level cooccurrence matrix (GLCM) is used to extract the texture features information. Then, this extracted color, and texture features are used to train the proposed CATDSNet using hybrid extreme learning machine (HELM) classifier for classifying various diseases such as healthy, appendicitis, bronchitis, gastritis, heart disease, and pancreatitis. The obtained simulation results on tongue image dataset disclose the superiority of proposed CATDSNet model as compared to state-of-the art approaches such as random forest, support vector machine (SVM), and SVM-based recursive feature elimination (SVM-RFE) with a classification accuracy of 92.3%, precision of 92.39%, recall of 92.06% and F1-score of 92.02%.

25

Authors: P. Muthulakshmi, M. Parveen

At present, health prediction in contemporary life set off very much indispensable. Big Data exploration plays a major contribution to predict subsequent status of health and offers outstanding health consequence to people. A lot of research is persisting on predictive analytics utilizing optimized machine learning techniques to disclose healthier decision making. Big Data analytics strengthens exceptional opening to predict future health condition from health criterions and bestow finest outcomes. We used Big Data Predictive Analytic model for heart disease prediction using Z-score Normalized Iterative African Buffalo Optimization (ZN-IABO). It fills the missing values in the database based on Z-score Normalized Data Pre-processing that standardize the scores on same scale by dividing score deviation by standard deviation. Z-score normalization model is suitable for huge data sets especially for big data. Next, Iterative African Buffalo Optimization based Feature Selection process is applied to the pre-processed data that addresses pre-mature convergence. In the simulation, the proposed ZN-IABO method is also compared with numerous well-known algorithms ZN-IABO (without optimization), imperialist competitive algorithm, and FODW. The result illustrates that this proposed algorithm is very competitive compared with the cardiovascular disease dataset for addressing the theoretical issue and superior to solving the real-world issue. The proposed ZN-IABO method (with optimization), achieves enhancement in the heart disease prediction accuracy by 17%, minimization of heart disease prediction time, error rate, and space complexity by 22%, 39%, and 37% as compared to the ZN-IABO (without optimization), imperialist competitive algorithm, FODW, MLBO respectively.

49

Authors: Arnika Jain, Rabins Porwal, Vineet Kansal

Agile development has proven its benefits over traditional software life cycle development models for more than a decade. Through offering agility as responsiveness to change, its practitioners maintain flexibility to accommodate change even at later stage of development process and simultaneously depict progress through iterative development. This scope creep and rapidly emerging market needs is placing more stringent constraints on agile projects in terms of meeting timelines, containing cost and adhering to quality. Though methodologies offered by agile have spawned several practices that are claimed to be efficacious, but still empirical validation and evolvement of them is required before adopting them as is for any project. This paper proposes and provides the design of a new methodology named as "Eclectic Agile Methodology". The key concept of eclectic methodology is to provide numerous advantages to mid and large scale organizations through enabling them to utilize novel and efficient ways of processes, tools and validations to assess the vast benefits before choosing any methodology upfront. The inspiration for designing this methodology is to address and overcome various untouched challenges in the current existing agile methodologies like but not limited to tightly coupling with upstream and downstream applications, sprint encountering frequent unplanned severity 1 issues which need immediate attention, shared hosted environments along with schema, dependencies outside the project, cross track impacts and distributed team.

79

Authors: San Pa Pa Aung, Win Pa Pa

Image Captioning (IC) is one of the most widely discussed topic in Artificial Intelligence. In this paper, Myanmar image caption is generated using EfficientNetB7 and Bidirectional Long Short-Term Memory (Bi-LSTM) with GloVe embedding and described about the comparative analysis results. For the purpose of achieving better performance, Myanmar image caption corpus is created and annotated over 50k sentences for 10k images, which are based on Flickr8k dataset and 2k images are selected from Flickr30k dataset. Two different types of segmentations such as word and syllable segmentation level are studied in text pre-processing step and then constructed our own GloVe vectors for both segmentations. As far as being aware and up to our knowledge, this is the first attempt of applying syllable and word vector features in neural network-based Myanmar IC system and then compared with one-hot encoding vectors on various different models. According to the evaluations results, EfficientNetB7 with Bi-LSTM using word and syllable GloVe embedding outperforms than EfficientNetB7 and Bi-LSTM with one-hot encoding, other neural networks such as Gated Recurrent Unit (GRU), Bidirectional Gated Recurrent Unit (Bi-GRU), and Long Short-Term Memory (LSTM), VGG16 with Bi-LSTM, NASNetLarge with Bi-LSTM models as well as baseline models. EffecientNetB7 with Bi-LSTM using GloVe vectors achieved the highest BLEU-4 score of 35.09%, 49.52% of ROUGE-L, 54.34% of ROUGE-SU4 and 21.3% of METEOR score on word vectors, and the highest BLEU-4 score of 46.2%, 65.62% of ROUGE-L, 68.43% of ROUGE-SU4 and 27.07% of METEOR score on syllable vectors.

100

Authors: Chegudi Ranga Rao, Ramadoss Balamurugan, RamaKoteswaraRao Alla

This paper presents an artificial rabbits optimization (ARO) based optimization methodology for the optimal location and rating of solar photovoltaic systems (SPVs) and passive power filers (PPFs) to mitigate the effects of non-fundamental frequencies due to the presence of non-linear and static loads in radial distribution networks (RDNs). The basic components of the distribution network are modelled for decoupled harmonic power flow (DHPF). The performance of RDN and harmonic levels is evaluated using DHPF. A multi-objective function using real power distribution loss (RPDL), voltage deviation index (VDI), total harmonic distortions in voltage (THDv), and current (THDi) is formulated by considering different equal and unequal planning and operational constraints. The effectiveness of the proposed methodology is evaluated on IEEE 33-bus RDN. Also, the computational efficiency of ARO is compared with other recent meta-heuristic approaches, namely the coyote optimization algorithm (COA), the butterfly optimization algorithm (BOA), the future search algorithm (FSA), and the pathfinder algorithm (PFA). The comparative analysis has highlighted the superiority of ARO in terms of global optima over the other compared algorithms. The reactive power losses and AVDI are reduced to 72.7865 kW, 50.6519 kW, and 0.0214 from 210.11 kW, 143.03 kVAr, and 0.0637, respectively. Also, integration of SPVs and single-tuned PPFs at optimal locations with appropriate sizes results in significantly improving the RDN power quality and performance significantly. The maximum in the entire network is observed at 5.21% of the fundamental frequency at bus-13. As per the IEEE 519 standards, it is required to maintain less than 5% and it is reduced to 3.98% with optimal PPF integration by the proposed method.

120

Authors: Eaint Mon Win, Si Si Mar Win

Community structure is one of the main structural features of networks and detecting overlapped community structure is an important field in social network analysis. In recent years, local community detection algorithms which detect overlapped community structure have been developed. However, the most existing algorithms suffer unstable community structure because the influences of parameter for controlling community’s resolution of fitness evaluation functions where used in identifying communities. Therefore, this article designs the optimized parameter evaluation formula to avoid the parameter influences and the algorithm is modelled on local expansion strategy. In this work, firstly identifies the seed or core node by using extended jaccard similarity and form initial community via seed. Then local community is detected by expanding the initial community with fitness function based on proposed optimized parameter evaluation and finally overlapped nodes are identified by merging detected local communities. In this article, the algorithm is implemented by using small dataset from network data repository site and large networks from Stanford large network datasets collection. The performance results of algorithm are compared with LFM (Local Fitness Method), OSLOM (Order Statistics Local Optimization Method), DEMON (Democratic Estimate of Modular Organization of a Network), NILPA (Node Important based Label Propagation) and GREESE (Greedy Coupled-seeds Expansion) algorithm. Then, the proposed algorithm proves that it effectively performs and saves execution time.

142

Authors: Yuni Yamasari, Anita Qoiriah, Naim Rochmawati, Kayo Yoshimoto, Rafif Aydin Ahmad, Oddy Virgantara Putra

The corona pandemic has changed learning methods from face-to-face to online. However, the application of online learning creates difficulties for teachers in monitoring student behavior because of the reduced direct interaction. This problem causes the learning process to be less optimal. Moreover, students may fail to achieve learning objectives. This research addresses this problem by building a model to detect student behavior in this online learning. It focuses on finding an optimal model by exploring the ensemble learning-stacking method based on a combination of SVM kernels (Linear, Polynomial, RBF, Sigmoid). After the model was built, it was evaluated using two performance measurement techniques, namely: cross-validation and percentage split, and several performance measures, namely: AUC, Accuracy, F1, Precision, and Recall. The evaluation results show the superiority of the models applying ensemble learning over those without it. In terms of accuracy, the highest result in the cross-validation technique is 98.4%, achieved by three models employing stacking. Those three are with base learners combination of linear-polynomial-sigmoid kernel (LinPolSig_Stack), a combination of linear-RBF-sigmoid kernel (LinRBFSig_Stack), and a combination of all kernels-linear, polynomial, RBF, sigmoid (AllKernels_Stack). In the percentage split technique, the highest performance is 97.4%, achieved by two models implementing ensemble learning-stacking with base-learners combination of RBF-sigmoid kernel (RBFSig_Stack) and combination of linear-polynomial-sigmoid kernel (LinPolSig_Stack). Finally, the highest performance of these models is equivalent to the minimum error in detecting student behavior. Detection errors were only three students in the three models in the cross-validation technique and only six in the two models in the percentage split technique.

163

Authors: S. V. Bharath Kumar Reddy, Chennupati Sumanth Kumar

In multiple input multiple output (MIMO) radar systems, the design and development of orthogonal frequency division multiplexing (OFDM) – linear frequency modulation (LFM) waveforms is one the most challenging and crucial tasks. Similar to this, increasing the pulse properties of spatially synthesised signals is also crucial in MIMO systems since it makes it possible to successfully eliminate grating sidelobes. Various joint optimization strategies are established in the various existing research works to enhance the overall performance of radar communication systems. Even yet, it has drawbacks including harder to understand computations, false detection rates, and reduced efficacy. Therefore, the proposed work intends to develop an integrated particle swarm optimization (PSO) – sequential quadratic programming (SQP) mechanism for improving the pulse compression properties and suppressing the grating sidelobes. Here, a modification is made to the designed LFM waveform to improve its frequency steps while maintaining balanced sidelobe and orthogonal features. By using the different pulse compression properties, the performance and outcomes of the proposed PSO-SQP based joint optimization model are evaluated and compared through simulation analysis. In this assessment, the average pulse compression, cross correlations, transmit beam patterns, and correlation properties are evaluated for varying spatial synthesis signals for the existing methods with the proposed method. Moreover, the estimated values of these parameters are compared with the conventional and proposed PSO and joint optimization techniques. The proposed method produced better results as compared with the existing methods with respect to various pulse compression properties such as auto-correlation side lobe peaks (ASP) and cross-correlation side lob peaks (CP). Based on the results, it is found that the proposed mechanism is capable of achieving the best ASP (19.4 dB), CP (14.02 dB) and ASLE (0.3745) among all, so assures the high range of resolution of the radar communication system.

187

Authors: Anandhavalli A, Bhuvaneswari A

Internet-of-Things (IoT) is one of the modern communication frameworks which is used to interconnect heterogeneous computational devices and to provide the expertise to exchange data among them without human-to- human or human-to-computer interdependence. This versatile heterogeneous communication is achieved by the unique identifier abstraction. Key exchange mechanisms are playing a vital role in IoT to procure the secured communication between the devices. The challenge in designing Key Exchange Mechanisms for IoT is increased due to the heterogeneous nature of the system. The IoT network can have some high-power computational rigs along with tiny low powered domestic appliances at the same time. The key exchange mechanism should be secure enough to protect the communication between the high computational power nodes and it should also be low power operational to deal with the low computational power devices. The proposed work of MLKEM formed the new Key Exchange Mechanism for IoT nodes based on Dual Rosernberg Pairing Location Masker and Fuzzy Miller's Elliptic Curve. The result of this proposed work is to exchange the new authentication key between the IoT nodes with secured manner. The new Key Exchange Mechanism are analysed by OPNET Network Simulator Tool. The introduction of Dual Rosenberg Pairing Location Mask is used to improve the security and Fuzzy Miller’s Elliptic Curve Key exchange is used to provide a power rational communication between the nodes in the IoT network.

214

Authors: N. D. S. S. Kiran Relangi, Aparna Chaparala, Radhika Sajja

In recent decades, the groundwater quality monitoring application gained more attention among the researcher community to assess the groundwater quality. The water quality index (WQI) is one of the effective models used for assessing the groundwater quality, which is not always superior in classifying the groundwater quality, especially in the large scale databases. Therefore, a new ensemble model is developed in this manuscript for classifying the ground-water quality. After collecting the data from the real-time and Indian water quality databases, the WQI calculation and the data denoising (Z-score and Min-Max normalization techniques) are accomplished. From the denoised data samples, the optimal features/attributes are chosen by implementing enhanced whale optimization algorithm (EWOA). Usually, the traditional WOA is computationally complex to explore the global solutions, therefore, a fitness function probabilitypro is included with the WOA for enhancing convergence speed and classification accuracy. The chosen optimal features/attributes are fed to the ensemble model: AlexNet and K-nearest neighbor (KNN) for classifying the types of groundwater quality. The introduced ensemble based EWOA model has achieved 99.88% and 99.98 (very near to 100%) of classification accuracy on the real time database and Indian water quality database.

235

Authors: Nitesh Sureja, Priya Patel, Mohit Rathod, Monika Labana

A crucial and important task in machine learning is feature selection (FS). The primary goal of the feature selection task is to minimize the dimension of the feature set while preserving performance accuracy. In order to address the FS task, a discrete moth flame algorithm that is combined with levy flights (DL-MFA) is presented in this research. The proposed DL-MFA imitates the natural navigational patterns of moths. The moths move along a straight line at a constant angle in the direction of the true light source (the moon) known as transverse orientation. Additionally, moths are drawn to artificial lights like fires and because of the close proximity; they constantly adjust their flying angles, creating a spiral path. In order to maintain healthy population diversity and increase the global search capabilities of the algorithm, the levy flight search technique is also used as a regulator of the moth position updating mechanism. The five swarm intelligence algorithms (SIAs) are contrasted with the proposed algorithm using measures such as entropy, purity, completeness score (CS), and homogeneity score (HS). For evaluating fitness, the SSE fitness function is utilised. The outcomes have shown that the proposed algorithm achieved purity values in the range 90% to 100%, and entropy 10% to 50%. Proposed DL-MFA has also achieved homogeneity score and completeness score up to 50%. These results prove that the proposed algorithm is better than its state-of-the-art competitors.

254

Authors: Pulung Nurtantio Andono, Guruh Fajar Shidik, Dwi Puji Prabowo, Dzuha Hening Yanuarsari, Yuslena Sari, Ricardus Anggi Pramunendar

Indonesia is home to several endangered species. This paper suggests a way for automatically categorizing bird sound patterns. This experiment used publicly available bird data to obtain bird sound patterns. However, bird sounds also contain noise, and the features are inappropriate, so data processing is required to reduce noise and select valuable features. YAMNet sound classification network data processing and feature selection incorporating Particle Swarm Optimization (PSO) may be used to decrease noise in bird voice data and choose appropriate features. All steps are completed before and after the gammatone frequency cepstral coefficient (GFCC) approach is used as feature extraction. This is ended by the KNN classification method to get the classification performance results. The experimental findings demonstrate that our proposal offers a performance for bird identification recognition of 78.33%, which is 1.27% better than our prior study. This finding outperforms previous research that solely used the dimensionality reduction-based classification algorithm, without picking the most important features.

277

Authors: Areej Abdulmunem Ahmed, Magda Mohamed Madboly, Shawkat Kamal Guirguis

The Cloud computing is a repository that stores massive amounts of data that is accessible online. There are many security and data privacy concerns threatening cloud computing services, especially through key sharing. So, there is an important need to improve security algorithms. An electrocardiogram (ECG) measures heart performance and detects arrhythmias. Nowadays, they are used for identification because they vary from one person to another. Therefore, securing an ECG signal is important. Homomorphic encryption (HE) is a security solution that processes encrypted data without decryption. HE supplies the same raw data results. In this paper, to enhance security and maintain data privacy while getting rid of sharing key restrictions, a new fully homomorphic encryption algorithm with an advanced encryption standard (FHEAES) is proposed to encode the ECG signals. This is accomplished by proposing the evaluation process to be appended to an AES algorithm as an additional security layer, which performs a variety of mathematical operations homomorphically on the encrypted data without decrypting it, instead of the original data. The evaluation process is proposed to be a linear polynomial function with variable values to enhance security. A Pan and Tompkins algorithm is used to process ECG signals. Then, the FHEAES algorithm was used for encoding the ECG signal. At last, ECG signals were classified for further diagnosis. According to the findings, FHEAES will take 1.078950 × 1033 years to break or hack. This is due to the using of the proposed three encryption keys to make it harder to break. Also, the FHEAES algorithm has 95.8% accuracy in classification and 100% accuracy in decryption (recovering the original signal). When compared to other algorithms, it is nearly 6 times and 3 times faster during encryption and decryption, respectively. This is because it is proposed to be based on computationally lightweight matrix operations, making it perfect to address the noise that arises from any FHE scheme. Thus, FHEAES provides superior security and privacy while computing and is solid against attacks.

301

Authors: Duraisamy Deenadayalan, Athiyappagounder Kangaiammal

The rapid growth of e-commerce has caused product overload where customers on the Web are no longer able to effectively choose the products they are exposed to. To overcome the product overload of online shoppers, a variety of recommendation methods have been developed. Collaborative Filtering (CF) is the most successful recommendation method, but its widespread use has exposed some well-known limitations, such as sparsity and scalability, which can lead to poor recommendations. This work proposes a recommendation methodology based on Web Usage Mining (WUM), and machine learning methods to enhance the recommendation quality and the system performance of current CF-based recommender systems. WUM populates the rating database by tracking learners’ behaviors on the Web, thereby leading to better quality recommendations. The data is collected from user profile and their preferences and also the weblink and usage. Based on the CF and WUM data, the recommendations are provided. The Random Forest (RF) and Extreme Gradient Boosting (XGBoost) classifiers is used to improve the performance of searching for nearest neighbors through Shuffled Frog Leaping Algorithm (SFLA). Experimental results show that the proposed model is effective and can enhance the performance of recommendation. Results show that the proposed SFLA-XGBoost has higher average sensitivity: rating <=15 for CF by 11.49% for RF, by 7.52% for XGBoost and by 4.58% for proposed SFLA-RF respectively.

328

Authors: Chaithra, Lingaraju Gowdru Malleshappa, Jagannatha Sreenivasaiah

Classification and analysis are improved factors for the real-time automation system. We need to analyze the attribute level in the area to predict the type of web page that can be predicted. This paper proposed a novel feature prediction model and the classification algorithm to estimate the attribute level of data and its probabilistic image to analyze the type of content that can be predicted on the page. This is to classify the different types of web pages which is better to perform the web page recommendation. For this process, dynamic lexical magnitude pattern (DLMP) system and correlated boltzmann machines (CBM) based classification model are used to analyze the attribute and image of the page area. The dataset consists of collections of attributes and images at various data samples for a page. In the DLMP-PGDT-based feature analysis method; the extract of the attribute and image in various texture patterns are analyzed and framed as the pattern for the given dataset. Then from that, an improved neural network architecture based on the block probability analysis is used to classify the data pattern to predict the class of web page according to the features of the dataset. This classification model assists were to recommend the content. The result analysis presents the comparison result of the proposed work with the web page classification technique like TrAdaBoost algorithm where proposed technique achieved accuracy of about 98%.

350

Authors: Hany Mohamed, Ahmed Hamza, Hesham Hefny

The internet of things (IoT) has gained great importance due to its applicability in various daily life applications and its flexible and scalable framework. The wide and spreading use of IoT in the last few years has attracted intruders, who were able to take advantage of the vulnerabilities of any IoT framework due to the absence of robust security protocols. This discourages current and probable investors. Out-of-date intrusion detection models are mainly developed to support information technology systems using built-in, predefined patterns or highly imbalanced datasets. Over the past decade, deep learning models have outperformed traditional machine learning models in attack detection tasks. The biggest challenge in detecting zero-day attacks is determining the best deep-learning classifier. Numerous research initiatives have combined ensemble learning to improve performance, avoid overfitting, and minimize errors. In this work, to address this research gap, we propose a new enhanced meta-learning ensemble deep learning model based on stacking that combines the baseline deep learning models using two tiers of meta-classifiers. Then, conducting several experiments on two recent huge-size different IoT benchmark datasets to evaluate the performance of the proposed model. Different baseline deep learning classifiers are trained for each dataset, and their performance is compared to the proposed ensemble model. The findings show that the proposed ensemble model significantly enhances the classification accuracy of baseline deep-learning models.

375

Authors: Satyabrata Patro, Jyotirmaya Mishra, Bhavani Sankar Panda

BC (Breast Cancer) is predominantly prevalent in women and is the leading cause of mortality due to cancers which can be reduced using mammography screenings. The use of CNNs (Convolution Neural Networks), a type of deep learning method has proven to be highly successful in image identifications. However, the quality of acquired mammographic images is found to be low while being used by detection models and hence this work proposed hybrid MLTs (Machine learning Techniques) for overcoming low-quality issues in the prediction of BCs. Initially, Statistical correlation analysis-based pre-processing is introduced for improving classifier performances followed by a hybrid model which predicts BCs effectively. This work also introduces a novel building block called Fuzzy Scoring based Resnets (Residual Networks) and CNNs called FS-Resnet CNNs for optimizing networks. The proposed FS-Resnet CNN model is computationally efficient, less sensitive to noises, and efficient in memory usage. Experimental results show that the proposed model achieves 95% accuracy ,95.45% precision, 93% recall rate, 94.21% f-measure and 18.113(S)-time complexity.

402

Authors: Alaa H. Sheer, Hazim G. Daway

A powerful magnetic field and high-frequency radio waves are used in magnetic resonance imaging (MRI) to create highly detailed images. By examining anomalies in the brain, spinal cord, muscles and liver, MRI can be utilised to identify specifics about soft tissues. MRI is very helpful in finding cancers in these tissues. Brain images from MRI often suffer from insufficient detail and contrast. This study focused on improving brain images from MRI by relying on c-mean fuzzy segmentation to enhance and preserve the detail of the images and applying adaptive histogram equalisation with the introduction of a smoothing median filter for reducing blurring and increasing contrast using the sigmoid function. The method was compared with several methods, such as local fractional entropy, Riesz fractional, dual illumination estimation, fuzzy logic based on the sigmoid membership function, fuzzy and spline based dynamic histogram equalisation and modified colour histogram equalisation, by calculating quality standards, namely, NIQE and BRISQUE. Results show that the method obtained the best results, namely, 5.4003 (NIQE) and 36.614 (BRISQUE) values.

419

Authors: Nattapong Jundang, Suchada Sitjongsataporn

This paper presents how to modify the proposed Discriminant Feature Trace Transform (DFTF) algorithm for object rotation. This proposed DFTF algorithm based on the Trace Transform (TF) domain can collect data for the formation of distinctive features and is suitable for predicting the rotational direction of an object with the supervised machine learning. TF domain is used after removing the background image that transformation properties can be generated a feature referred to the rotation features of objects inside. After that, the proposed DFTF algorithm transforms the two-dimensional data of each image from the TF domain into the one-dimensional data that can indicate the direction of object rotation within the image. The results of the DFTF algorithm, which is a 1D data vector from each image, are then generated into the labeled datasets for machine learning algorithms including with the Naïve Bayes (NB) for predicting the rotation direction and Random Forest (RF) to reinforce the predicted values from the NB in the form of quadrant in which the interested object is rotated. The simulation experiments are conducted with two types of databases. The first experiment combines with the three different datasets which the results are very effective and can provide the average accuracy rate for all databases up to 99.8%. The second experiment is derived from the visualization of the water bottle production line. It measures the accuracy of the bottled water transfer to the second station. The proposed approach is an accuracy rate up to 92.2%.

460

Authors: Erwin Sutanto, Teguh Wijaya Purwanto, Fahmi Fahmi, Muhammad Yazid, Wervyan Shalannanda, Muhammad Aziz

Epilepsy patients might need continuous electroencephalography (EEG) monitoring to help them understand their own condition’s improvements or their doctors to determine the frequency of the seizures. In this study, we demonstrated how a low-cost, portable EEG headband could be used to detect absence seizures in epilepsy patients. The method used is Support Vector Machine (SVM) to separate the initial limit and Machine Learning with Tensorflow to predict its confidence level. Then, we tried to test our method on 2 other patients to see its measurement divergence. Furthermore, the Tensorflow library has been applied and developed to train and classify the data, which is also one of the novelty approaches in this study. From the result, closed eyes can be detected from open-eye conditions with more than 96 % accuracy and a loss of only 2.35%. The findings demonstrate the feasibility of detecting absence seizures using only two electrodes which were TP9 and TP10 of Muse headband which is also positioned in the ear like an ear-EEG. Overall, the study successfully developed a novel method utilizing a low-cost headband to provide affordable health system access.

484

Authors: Pramod Mutalik, Venkangouda C. Patil

A vehicular Ad hoc Network (VANET) is generally a heterogeneous wireless network generated between vehicles. The vehicles of the VANET have wireless transceivers and computerized control for allowing the vehicles to operate as network nodes, hence the vehicles can communicate in a VANET environment. However, the communication of VANET is affected because of the higher network congestion and energy usage caused by the dynamic topology of VANET. Therefore, an effective Traffic-Aware Routing (TAR) approach is required to be developed to enhance communication. In this paper, the Multi-Objective Delay Centric Enhanced Artificial Ecosystem-based Optimization (MDCEAEO) is proposed to develop a TAR in VANET. The End to End Delay (EED) is considered as primary cost in the MDCEAEO to develop the TAR where a vehicle’s average predicted speed is utilized for identifying the traffic in VANET. The developed MDCEAEO-TAR method is used to improve data transmission by avoiding collisions. The performances of MDCEAEO-TAR are evaluated using EED, energy consumption, Packet Delivery Ratio (PDR), and the routing overhead. The existing research such as LARgeoOPT, DREAMgeoOPT, ZRPgeoOPT, Improved Harmony Search (IHS) and Enhanced Distance, Residual energy based Congestion Aware Ant Colony Optimization (EDR-CAACO), Artificial Ecosystem-based Optimization (AEO), Fixed Step Average and Subtraction Based Optimizer (FS-ASBO) and Three Influential Members Based Optimizer (TIMBO) are used to evaluate the MDCEAEO-TAR. The PDR of the MDCEAEO-TAR is 0.9836 at 1000s, which is high when compared to the IHS.

502

Authors: Bouchaib Zohal, Houssam Amiry, Said Yadir, Elhadi Baghaz, Rachid Bendaoud, Abderrahim El-Abidi, Fatima Chanaa, Mohammadi Benhmida

In this paper, a study on photovoltaic module physical parameters behavior depending on irradiance G and module temperature TM is presented. From this investigation, new equations translating physical parameters behavior depending on TM and G are developed. The new proposed laws, which combine the effects of G and TM in a correlated manner, have the advantage of incorporating G effect on the temperature coefficients by including specific terms. The developed laws form a basis for predicting PV module performance and can be incorporated into maximum power point tracking (MPPT) programs to extract peak power output under different operating conditions. This study is conducted by examining the electrical performances at the maximum power (Im, Vm), the open circuit (0, VOC), and the short circuit (ISC, 0) points measured using an elaborate experimental platform. The photovoltaic module used is a monocrystalline type exposed to four G values of 700, 800, 880 and 1000 W.m-2 for temperatures TM between 16°C and 48°C.

532

Authors: Rana Saad Mohammed

The design of a cipher system is essential and required with the development of the technology era. The internet of things concept is modern technology requiring lightweight, secure algorithms to investigate the requirement of the light device. Furthermore, IoT devices have limited resources (IoT devices should minimize battery life, limited memory size, and response time). A stream cipher system is faster than a block cipher system and has an easy feature with hardware and software implementation. This paper designs a lightweight authentication encryption algorithm based on stream cipher and chaotic maps with sponge structure for IoT applications. Currently, the suggested approach uses text data for encryption and authorization. Additionally, the suggested algorithm will be able to use it with image data in the future. Analysis results of the proposed algorithm based on NIST randomness tests, execution time, memory space, and functional features. The results are compared with other previous works which used sponge structure (e.g. Ascon, Elephant, ISAP, Photon-Beetle, and Xoodyak, etc.). Randomness tests show that the proposed system is random and secure with investigate all the functional features except the parallelization feature which will work it in the future. And also it has good fast speed and less memory space.

561

Authors: Irmalia Suryani Faradisa, Oddy Virgantara Putra, Tri Arief Sardjono, Mauridhi Hery Purnomo

High-risk maternal health condition is alarming, especially in developing countries. Intensive monitoring is mandatory to prevent such issue. However, the long-term invasive method to pregnant women harms both the baby and the mother. In this research, we proposed a cost-efficient non-invasive foetal heartbeat classification based on a phonocardiograph with feature assembly. Since the high number of features and computationally expensive, we cut the size to half by utilizing Principal Component Analysis. Furthermore, data balancing using SMOTE is incorporated to improve classification performance. We proposed a method based on a neural network and optimized it using Random Search optimization. Eventually, the proposed method gained the top position in all data balancing compared to other machine learning algorithms, with 91.7 % for both accuracy and Area Under Curve with a score at 91.6 %.

584

Authors: Mahmoud S. Sayed, Ahmed A. A. Gad-Elrab, Khaled A. Fathy, K. R. Raslan

Content-based image retrieval (CBIR) is a popular approach for searching and retrieving digital images from labeled or unlabeled image collections using image content features. The most important parts of the CBIR system are the computational complexity and the retrieval accuracy. Many studies have been conducted to increase the accuracy of image retrieval systems. In this paper, we propose a new fully unsupervised content-based image retrieval (CBIR) technique to increase the efficacy of image retrieval using a clustering approach on unlabeled image dataset. The proposed method combines features extracted from both pre-trained AlexNet model and pulse-coupled neural networks (PCNN) to extract high and low level features. Then principal component analysis (PCA) is performed on AlexNet’s features and these combinations are fed to the K-means algorithm after normalization process. Then Euclidean distance is used to measure the similarity between query and stored images within the same cluster. Finally top similar images are ranked and retrieved. Experimental results on the benchmark Corel-1k and Corel-10k datasets show that the proposed method achieves high precision values of 92.8% and 64.8% respectively, on the top 20 retrieval levels compared to other methods.