International Journal of Intelligent Engineering and Systems

1

Authors: Nanang Triagung Edi Hermawan, Edi Winarko, Ahmad Ashari, Yus Rusdian Akhmad

Radiation monitoring data is very important to support the safety of nuclear installation. The data should be ensured generated and supplied by an authenticated sensor. The new sensor should apply mutual authentication to server before joint the system. This research proposes a high secure authentication protocol based on Eight Prime Number of Modified RSA Algorithm (EPNR) method as certificateless mechanism to protect authenticity and confidentiality of sensor’s identity. The main advantages of the protocol are designed for simpler direct point to point connection, using one-time random keys, and implementing five-barrier data protection mechanism. Its performances in context security, running time, and Quality of Services have been compared to other models. The proposed protocol is faster 29.9 times in random key generation, 7.27 times in encryption, and 11.1 times in decryption duration, compared to the others model based on RSA scheme. It has better performances in throughput and delay time. It also has enhanced resistance to statistical, replay, and MITM attack.

25

Authors: Divya Jegatheesan, Chandrasekar Arumugam

The intelligent transport system aims to control congestion and enhance the driving experience through a variety of technologies and communication systems. They provide us a lot of data which can be fed to machine learning technologies to provide further enhanced services to the common public as well as the transport officials. In this busy modern world, the usage of vehicles is highly increased and thus monitoring and detecting these vehicles onboard with better accuracy and less time cost is quite a challenging task. We propose an intelligent traffic management solution to detect vehicles (static and onboard) which finds its set-about in the field of machine vision technology called YOLO D-NET (You Only Look Once Dilated Net). YOLO D-NET puts forward a novel architecture which incorporates the implementation of traditional YOLO model alongside of dilated Convolution Neural Network (CNN). Our paper also focuses on making a constructive comparison between the four different models namely SSD (Single Shot Multibox Detector) Inception, Faster R-CNN (Region-based CNN), YOLO ResNet (Residual Neural Network) and the proposed model YOLO D-NET. We have used the predefined COCO (Common Objects in Context) dataset and the custom dataset to train the traditional models. Also YOLO D-NET makes use of the predefined YOLO weights followed by the dilated CNN layers. Our proposed model was found to be 97.5 percent accurate with enhanced precision and accuracy than the other models particularly with less training time than the traditional YOLO model.

46

Authors: Quyen Thi Nguyen, Minh-Phung Bui

This paper presents a new method based on Equilibrium Optimizer (EO) algorithm that is inspired from the mass balance of a control volume for traveling salesman problem (TSP). For enhancing the efficiency of EO, the 2-opt movement algorithm is used to update the solution generated by EO. The efficiency of the proposed EO for the TSP problem has been compared with Particle Swarm Optimization (PSO) and Genetic Algorithm (GA) on different instances consisting of the 14-city, 30-city, 48-city and 52-city. The calculated results show that for the large scale instances such as 48-city and 52-city, EO has found the better tour than PSO. In comparison with GA, EO has ability finding the best tour with the smaller mean and standard deviation. The comparisons with previous methods in literature have also demonstrated that EO has ability to search the better tour than other methods. Thus, the proposed EO can be a potential method for the TSP problem.

67

Authors: Khin Hsu Wai, May The` Yu

Effective and automated lane marking detection system is one of the important part of many intelligent transportation systems. Many researches of lane detection techniques have been conducted on academic and automobile industry and applied in intelligent transportation systems. However, It is still a challenging problem because of the various environmental conditions such as shadows, lane occlusion and illumination invariance that can decrease the lane detection performance. To deal with these issues, a reliable lane marking detection system that use the block-wise Hough based features is proposed. We propose the Region of interest extraction that divides the image into sub-image blocks to detect the lane line and block-wise Hough Transform feature extraction approach to extract the lane line features. Artificial Neural Network (ANN) applys the extracted features to detect the lane. The efficiency and accuracy of our system is evaluated by using the Caltech lane dataset. The average correct detection rate is 94.67% and the computational time is 17.64 ms/frame by applying the proposed method.

88

Authors: Azeddine Loulijat, Najib Ababssi, Makhad Mohamed

The main problem of the Doubly-Fed Induction Generator (DFIG) is its sensitivity to network disturbances especially voltage dip due to its stator windings being directly linked to the network. A voltage dip in the network results from high current peaks in the stator windings, since the stator and rotor are electromagnetically coupled, so these peaks provoke inrush current hard at the delicate back-to-back converters and an overcharge of the DC-LINK capacitor. In this perspective, several researchers are realized a linear and nonlinear controller with a passive protection method (PPM). The objective of this paper is to develop modelling and control models based on a true DFIG model considering the effect of all the parameters of a stator winding. The Second- Order Sliding Mode controller (SOSMC) is the non-linear control strategy used to control the dynamics of the DFIG through the rotor-side and network-side power converters under normal network conditions. When a voltage dip is detected, the Kalman observer (KO) contributes to a non-linear controller and the latter is coupled to a passive protection method that contains crowbar and DC-chopper circuits to improve the dynamics of the DFIG system to cross the network fault. Also, a comparative study is carried out between the developed controller and other linear ,non-linear approaches exist in the literature. The results obtained, under a maximum voltage dip (100%) on the 60 kV network, confirm its robustness and superiority in terms of limiting the DFIG graders at the beginning and on termination of a fault. In terms of Ir and Vdc values during a voltage dip, our control strategy does not exceed 1.45p.u and 1.2p.u, on the contrary, the ABC with HGO which is developed recently reaches 2.2p.u and 0.4p.u.

112

Authors: Yosepu Cooli, Chitraivel Mahesh

The development of computing technology has led to enormous improvement in the medical data and machine learning processing techniques. The increasing data usability in various field leads to the term big data, which is required in various applications such as medical, finance and so on. The existing models such as Random Forest, Support Vector Machine (SVM) and Decision tree have low capacity in handling the large volume of the data. In this research, the Recursive – Parallel Random Forest is proposed to effectively classify the large volume of medical data. The recursive partition method has been used to segment the input data and process the data in segmented manner. The recursive partition method segment the data and subset features based on data centric to improve the learning and provide adaptive gain ratio value to build Random Forest. The recursive partition method provides the data with high correlation that supports the random forest to perform the Random Forest parallel. The partition the data in data centric and balance the classes that helps the proposed model to handle the imbalance data effectively. The Parallel Random Forest method has been used to improve the performance of the medical data classification and the performance of Recursive partition- Parallel Random Forest (R-PRF) method was estimated using three UCI medical dataset such as Chronic Kidney Disease (CKD), Heart Disease and Diabetes. The standard classifiers and state-of-art method has been used to compare with the proposed R-PRF method and the experimental result shows that the R-PRF has the accuracy of 92.01 % and the existing Improved SVM radial method has 89.9 % accuracy.

129

Authors: Chinthakunta Manjunath, Balamurugan Marimuthu, Bikramaditya Ghosh

Equity market forecasting is difficult due to the high explosive nature of stock data and its impact on investor's stock investment and finance. The stock market serves as an indicator for forecasting the growth of the economy. Because of the nonlinear nature, it becomes a difficult job to predict the equity market. But the use of different methods of deep learning has become a vital source of prediction. These approaches employ time-series stock data for deep learning algorithm training and help to predict their future behavior. In this research, deep learning methods are evaluated on the India NIFTY 50 index, a benchmark Indian equity market, by performing a technical data augmentation approach. This paper presents a Recurrent Neural Network (RNN), Long Short Term Memory (LSTM), and the three variants of Gated Recurrent Unit (GRU) to analyze the model results. The proposed three GRU variants technique is evaluated on two sets of technical indicator datasets of the NIFTY 50 index (namely TA1 and TA2) and compared to the RNN and LSTM models. The experimental outcomes show that the GRU variant1 (GRU1) with TA1 provided the lowest value of Mean Square Error (MSE=0.023) and Root Mean Square Error (RMSE= 0.152) compared with existing methods. In conclusion, the NIFTY 50 index experiments with technical indicator datasetTA1 were more efficient by GRU. Hence, TA1 can be used to construct a robust predictive model in forecasting the stock index movements.

154

Authors: Srinivasulu Sirisala, S. Rama Krishna

The successful routing in Mobile Ad hoc Networks (MANETs) completely depends on the cooperation established between the mobile nodes during data dissemination. The resource constraint feature of mobile nodes is the main reason behind its non-cooperativeness attributed towards other interacting nodes of the network. This degree of non-cooperation rendered by the mobile node disturbs the routing and degrades the network performance. Thus, a node cooperation enforcing approach that distinguishes genuine node from malicious node is essential for sustaining the network performance. In this paper, Gaussian Trust Factor-based Grey Decision-Making Technique (GTF-GDMT) is proposed for maintaining the network performance independent to the influence of non-cooperative malicious nodes. This proposed GTF-GDMT identified the malicious nodes based on the trust factor that integrates the parameters of link stability and node reputation determined through direct and indirect monitoring process. This computation of trust is achieved using the Gaussian probabilistic distribution model that utilizes multi-perspective parameters that forms the input to the decision-making process. It is proposed for attaining the trade-off between energy consumption, transmission performance and security. The simulation experiments of the proposed GTF-GDMT conducted using ns-2 portrayed its excellence over the baseline schemes in terms of throughput, packet delivery rate, packet delay, energy consumptions and packet drop independent to the number of mobile nodes and malicious nodes in the network.

177

Authors: Suhariyanto, Rachmad Abdullah, Riyanarto Sarno, Chastine Fatichah

Sentiment analysis can provide rough recommendations in the form of sentiment from a collection of reviews or can provide recommendations in more detail about sentiment in a particular aspect called aspect-based sentiment analysis (ABSA). Sentiment analysis based on many aspects has been carried out but its accuracy is still being developed. In previous research, most research was carried out on explicit and implicit aspects and opinions and was carried out in simple sentences. The purpose of this research is to analyze the sentiment of restaurant reviews using the rule grammar method to extract implicit aspects - explicit opinions in four sentence models, namely simple (Si-AIOE), compound (Co-AIOE), complex (Ce-AIOE), and compound-complex (CoCe-AIOE). The ABSA method is proposed using the development of a grammatical rule extraction method to extract explicit and implicit aspect words and opinion words as the basis for sentence extraction. Rules making is done to take explicit and implicit aspect words and opinion words in Si-AIOE, Co-AIOE, Ce-AIOE, and CoCe-AIOE so that the comparison of the evaluation values can be known. This research uses the Semeval 2015 dataset on Restaurant reviews from the Tripadvisor Website which has been annotated as sentence data for ABSA. The aspect categorization process is then used to categorize aspects into 4 aspect categories, namely Ambience, Food, Service, and Price using hybrid approach. The hybrid approach is combined using Elmo-Wikipedia, grammatical rule extraction, WordNet, TF-ICF, and semantic similarity methods. The results of the aspect extraction obtained value of precision, recall, and f1-measure of 0.80, 0.84, and 0.82, respectively. Meanwhile, the ABSA process uses SentiCircle to classify sentiments into two, namely positive and negative. The results of the ABSA showed that the performance of proposed method achieve for precision, recall, and f1-measure were 0.84, 0.89, and 0.87, respectively.

199

Authors: Reza Fauzi Iskandar, Edi Leksono, Endra Joelianto

This paper presents a control strategy of maximum power point tracking and focuses on hybrid q-learning and type-2 fuzzy logic control strategy. Photovoltaics have non-linear voltage and current characteristics which influenced by temperature and exposure of solar irradiation so that the maximum power point (MPP) can change at any time. Unfortunately, operating conditions outside the MPP could reduce the efficiency of electrical power transfer from the photovoltaic system to the load. Q-learning (QL) has the potential to provide action decisions of duty cycle signal based on the state of the power gradient percentage range. The Q-learning hybrid method with Type-2 fuzzy logic control is proposed to provide correction of QL control signal by considering the uncertainty direction of power point shifting. The proposed strategy is important particularly to avoid a sudden change in solar irradiation exposure. The simulation results show that the Q-learning hybrid type-2 fuzzy logic control based MPPT response in the photovoltaic control system has tracking efficiency about 97%, rapid rise time of 0.08s, settling time of 0.23s, low overshoot and stable response for handling the change of irradiation and temperature exposure simultaneously.

226

Authors: Manjunath Hanumanthappa Rangappa, Guruprakash Chirathahalli Dyamanna

The combination of both static sensors and mobile sensors is called as Hybrid Wireless Sensor Network (HWSN). Static sensors monitor the environment and report events occurring in the sensing field. Mobile sensors are then dispatched to visit these event locations to conduct more advanced analysis. These sensor nodes have major energy constraints as they are small and their battery can’t be replaced and collaborate together in order to gather, transmit/forward the sensed data to the base station. Consequently, data transmission is one of the biggest reasons for energy depletion in HWSN. Clustering is one of the most effective techniques for energy efficient data transmission in HWSN. This paper proposes an Adaptive Hybrid Cuckoo Search (AHCS) algorithm. AHCS improves the Lévy flight method and population evolution strategy of the cuckoo search algorithm, and introduced a mutation operation operator. Inspired by the idea of position update of Grey Wolf Optimization (GWO) algorithm, this paper introduces the inertia weight w in the Lévy flight method of CS algorithm, and gives the new dynamic adjustment methods of parameters α and β respectively. The proposed algorithm is evaluated in terms of response time, packet drop ratio, energy consumption, end to end delay and dead time. Furthermore, the performance of AHCS-GWO is evaluated with existing methods such as classical algorithms, Evolutionary Multipath Energy-Efficient Routing protocol (EMEER) and Yellow Saddle Goatfish Algorithm (YSGA). The residual energy of the AHCS-GWO method for 200 nodes is 0.57J, it is high when compared to the existing methods.

247

Authors: VietHung Nguyen, DinhToan Nguyen, NhatTan Nguyen

The trend of Industry 4.0 development in the maintenance field requires the big data analysis of system operation for predicting, or identifying any possible problem. This study proposed an automatic fault diagnosis method for rolling bearing. Ensemble empirical mode decomposition (EEMD) is used to preprocess the vibration signals of rolling bearing which solved the problem of mode mixing existing in empirical mode decomposition (EMD). The features are then extracted from the first IMFs in the time and frequency domain to form an initial fault feature of the complex vibration signal. The Laplacian score (LaS) method for selecting features is then utilized to construct a new sensitive feature according to their significances, named EEMD-LaS. An optimized pattern classification method is constructed to achieve an automatic diagnosis of bearing fault. This method uses a backtracking search optimal (BSO) algorithm to optimize the parameters of the least-squares support vector machine (LSSVM) classifier model, named BSO-LSSVM. Finally, the EEMD-LaS and BSO-LSSVM base proposed method for bearing fault diagnosis is applied to analyze the experimental data. The result indicates that the proposed EEMD-LaS – BSO-LSSVM method can effectively diagnose the fault categories with 100% accuracy of the outer-race fault and normal bearing condition; and 97% minimal accuracy of the severe conditions of the inner-race fault and roller fault of rolling bearing.

269

Authors: Awan Uji Krismanto, Indra Soegiarto, Abraham Lomi, Irrine Budi Sulistiawati, Herlambang Setiadi, Muhammad Abdillah

One of the main focus in integrating large scale wind power plant is how to maintain voltage stability under different power injection from wind farm. In this paper, effects of large-scale wind power plants on voltage profile of power system is investigated. Practical test system of South-West Sulawesi, Indonesia with integration of two large scale wind power plants are considered. It was monitored that the increasing wind enhanced the voltage profile of the system. Impacts of wind power integration on power system voltage profile were investigated in this paper. Probability analaysis based on MCS were conduted to observe the impacts of uncertain power injection from wind farm on voltage fluctuation. It was clearly monitored that the probability distribution of bus voltage varied accordingly depending to location and capacity of wind farm. It was also monitored that enhancement of voltage profiles increased in proportion with power injection from wind farm. Thus, having more power production from wind farm results in better load-ability and eventually improved voltage stability condition of power system as shown in bus 31 (the voltage magnitude increased from 0.942 to 0.952 pu).

294

Authors: Rachmad Abdullah, Suhariyanto, Riyanarto Sarno

TripAdvisor is one of the most popular e-commerce platforms in the tourism sector in Indonesia. TripAdvisor give Traveler Choice Award every year in Indonesia through user reviews. However, online text-based reviews are often associated only with evaluation scores that do not pay attention to the context and meaningful content of the review itself, either explicitly or implicitly. Moreover, the sentence structure of the review can have an impact on the goal of the target sentiment which is nothing but an aspect of the review itself. This research discusses aspect based sentiment analysis for explicit and implicit aspects. This research starting with taking the TripAdvisor website restaurant product review dataset to measure customer satisfaction based on four aspect categories of Ambience, Food, Service, and Price. Furthermore, the aspect word extraction and opinion word extraction processes in the case of explicit sentences for simple, compound, complex, and compound-complex sentence structures are carried out using grammatical rule extraction. This research also works on implicit sentence cases for simple sentence structures. Aspect categorization process uses hybrid approach. Aspect and opinion keyword extraction process uses the ELMo-Wikipedia. Then WordNet and TF-ICF are used to expand the meaning of aspect and opinion that has been taken. The last stage is the aspect based sentiment analysis process, both explicit and implicit sentences using SentiCircle. This research can produce two evaluations of sentiment classification, namely positive and negative. The results of the aspect extraction obtained an evaluation of the aspect categorization for each precision 0.82, recall 0.87, and f1-measure 0.86. Meanwhile, the results of the sentiment analysis showed that the respective evaluations for precision 0.87, recall 0.92, and f1-measure 0.89.

320

Authors: Antono Adhi, Budi Santosa, Nurhadi Siswanto

Studies and research related to the Vehicle Routing Problem (VRP) are mostly carried out on simple constraints. The study was conducted to find the shortest route that the carrier will take. Effective algorithms are sought to find better solutions. Another study will develop VRP cases to be closer to real conditions. In actual conditions, the VRP cases are very complex. Several solutions that are considered good can be implemented to solve this problem. VRP in shipping bulk products, such as oil, cements, and fertilizers is one of the daily logistical cases carried out in an archipelago country. Bulk products are shipped from the depot port city to other islands. The optimum solution is needed to make the delivery shorter distance and the products can be sent faster. The solution to the VRP case with a single product can no longer solve the problem of bulk product shipment. Bulk product shipment involves more than one product with more than one ship compartment. The number of ship compartments that are smaller than the number of products will further increase the complexity of the VRP of bulk product shipment. The combination of several metaheuristic methods and strategies of product selection to fill undedicated compartment of ship is used to find the shortest route. Modified Nawas Enscore Ham (M-NEH), Particle Swarm Optimization (PSO), and 3-Opt algorithms are combined to solve VRP. This hybrid algorithm called Hybrid Particle Swarm Optimization (HPSO) is improved and supports each other. HPSO provides a better solution than those algorithms separately and other metaheuristics namely Tabu Search (TS) and Genetic Algorithm (GA). The implementation of the algorithm to solve real data of bulk cement transportation supports the conclusion that HPSO is superior to the other algorithms. It has 0.99% more effective than the second best and about 3.42% more effective than the average total distance of five other compared metaheuristic methods.

348

Authors: Hussein A. A. Al-Khamees, Nabeel Al-A'araji, Eman S. Al-Shamery

Many different applications in the real world can generate huge amount of data, that has unconventional features including massive size, fast access, besides the evolving in its nature; this is data stream. Data stream clustering algorithms began to grow at breakneck speed. evolving Cauchy (eCauchy) is a significant algorithm of density-based data stream clustering. The major limitation of eCauchy is the high number of clusters generated in dynamic environments. This paper presents an evolving model for data stream by optimizing e-Cauchy algorithm to decrease the number of clusters and reach to an ideal number by implementing evolving mechanisms (adding, merging, splitting clusters) based on a specific membership function. Model is tested by two real datasets NSL-KDD99 and keystroke. Proposed model outperforms two other algorithms, e-Cauchy and FEAC-Stream. Model constructs five and four clusters with less time to implement 1.30 and 2.30 minutes respectively for each dataset.

369

Authors: Seyed Alireza Bashiri Mosavi, Seyed Yashar Banihashem

Monitoring transient excursions of power system dynamic response is the most significant task for a reliable power supply. Hence, pursuing the best-laid features alongside a fit learning model for precise mining on the inherent characteristic of fast transient phenomena is considered a great concern on transient stability analysis (TSA). To this end, first, we discussed the root cause involved in selecting the cross-relevance (CR) features that influenced TSA. Furthermore, we regarded the geometric form of CR (GCR) features to measure the partial-whole manner of power system stability limits. Next, we triangulated Python technology, PSS/E, and MATLAB for generating the GCR multivariate trajectories. In the learning phase, to shroud the linear alignment weakness, the elastic kernel plugged into the support vector machine is considered for transient stability prediction (TSP). The results show an average accuracy near 97% with an acceptable average processing time of 691.7 milliseconds on test cases for TSP.

395

Authors: S. Beghin Bose, S. S. Sujatha

Cloud computing is the on-demand availability of internet-based computing services, especially software, large amounts of data storage, operating systems, and other computing resources. Service Level Agreement (SLA) violation is the most critical problem in cloud computing. SLA violation creates many problems for cloud service providers and cloud customers. Due to this, cloud customer gets low-quality cloud service. Thus, designing an effective cloud service recommendation algorithm is a critical research problem in cloud computing. The primary objective of this research is to determine the optimal cloud service from functionally equivalent cloud services that better fit the user's requirements (latency, throughput, response time, and cost). The efficiency of cloud services varies according to the time and location of the virtual machine. First, this proposed method determines the correlation between active user requirements and cloud services. Second, strongly correlated services are separated into two clusters based on the virtual machine's location and the cloud service's data transmission rate. For this purpose, two lightweight clustering algorithms have been proposed. A modified multilayer perceptron algorithm has been developed to recommend the optimal cloud service to the active user from the two clusters. The open-source WS-Dream dataset is used to train and validate the proposed MLP. The training efficiency, prediction accuracy, and performance of the proposed MLP-based cloud service recommendation system are experimentally compared to the existing cloud service recommendation systems analysed in the literature study [20, 2, 22, 23]. Compared to existing cloud service recommendation approaches, the MAE and RMSE values of the proposed cloud service recommendation system are less than one. In terms of accuracy, the suggested method obtains a precession of 94 %, a recall of 97 %, and an F1-Measure of 96 %, all of these are significantly better than the existing cloud service recommendation methods. Finally, experimental results prove that the overall performance of the proposed method's throughput (increase 10 MBPS), latency (reduce 10 ms), the response time (reduce 17 ms) and service recommendation time (reduce 5ms) is more robust than existing methods.

422

Authors: Jai Bhagwan, Sanjeev Kumar

Cloud computing technology is becoming popular in both academia and industries these days because of its dynamic and flexible infrastructure which provides good computing facilities through web networks without a place specification. In order to process the workflows, the cloud technology uses datacenters which consist of hosts having various homogeneous and heterogeneous virtual machines. Virtual machines play the role of real machines with multiple operating systems environments. Due to a wide acceptance of the cloud, the load on the cloud is increasing regularly. So, scheduling plays a key role to manage a machine load by mapping several workflows with available virtual machines. The scheduling can be divided into two types: dynamic and static scheduling. Dynamic scheduling can be done effectively by meta-heuristic computing techniques. In the field of cloud computing, many scientists have developed various algorithms for workflow scheduling in order to schedule the workflows. The Cat Swarm Optimization works fine as compared to the Max-Min algorithm, Particle Swarm, and Ant Colony Optimization algorithms. In this research, a new algorithm is designed named H-CSO by using the concept of Heterogeneous Earliest Finish Time and Cat Swarm Optimization algorithms. After experiments, it is found that the proposed H-CSO algorithm gives efficient makespan at realistic costs as matched to the Cat Swarm Optimization. The newly designed H-CSO algorithm is 2.99%, 2.87%, 3.35%, and 5.77% efficient for CyberShake_1000, Montage_1000, Inspiral_1000, and Sipht_1000 datasets respectively as compared to the standard Cat Swarm Optimization in terms of average makespan reduction. In case of cost reduction, the proposed algorithm is 4.03%, 5.12%, 2.42%, and 3.93 effective as compared to the Cat Swarm Optimization for CyberShake_1000, Montage_1000, Inspiral_1000, and Sipht_1000 datasets respectively.

446

Authors: Lindung Parningotan Manik, Zaenal Akbar, Hani Febri Mustika, Ariani Indrawati, Dwi Setyo Rini, Agusdin Dharma Fefirenta, Tutie Djarwaningsih

Knowledge-based chatbot (KBC) has grown in popularity in recent years and has been widely used for various use cases. Building KBC from scratch using deep learning (DL) is challenging since no prior historical data exists. Meanwhile, DL systems need a vast Volume of data to be trained. This paper proposes a novel framework to create an intent classifier of the KBC used to detect in-scope (IS) and out-of-scope (OOS) intents. We introduce an automated queries generator to create IS intents employed as the training data from an ontology input. We utilize Bidirectional Encode Representations from Transformers (BERT) fine-tuning as the backbone of our DL system. Moreover, we present a Bayesian approach as an extension of the BERT to classify OOS queries with minimal OOS training data. The experiments result show that the proposed method manages to achieve an F1 score of 100% for IS intents and 86% for OOS queries.

469

Authors: Ali Saeed Alfoudi, Ali Hakem Alsaeedi, Mohammed Hamzah Abed, Abayomi M. Otebolaku, Yasmeen Sajid Razooqi

Palm Vein Identification (PVI) is a modern biometric security technique used for enhancing security and authentication systems. The key characteristics of palm vein patterns include its uniqueness to each individual, its unforgettability, non-intrusiveness and its ability for disallowing unauthorized persons. However, the extracted features from the palm vein patterns are huge with high redundancy. In this paper, we propose a combined model of two-Dimensional Discrete Wavelet Transform, Principal Component Analysis (PCA), and Particle Swarm Optimization (PSO) (2D-DWTPP) that feeds wrapper model with an optimal subset of features to enhance the prediction accuracy of -palm vein patterns. The 2D-DWT extract features from palm vein images, using the PCA to reduce the redundancy in palm vein features. The system has been trained to select high recognition features based on the wrapper model. The proposed system uses four classifiers as an objective function to determine PVI which include Support Vector Machine (SVM), K Nearest Neighbor (KNN), Decision Tree (DT) and Naïve Bayes (NB). The empirical results proved that the proposed model has the best results with SVM. Moreover, our proposed 2D-DWTPP model has been evaluated and the results show remarkable efficiency in comparison with AlexNet and other classifiers without feature selection. Experimentally, the proposed model has better accuracy as reflected by 98.65% whereas AlexNet has 63.5% accuracy and the classifier without feature selection process has 78.79% accuracy.

492

Authors: Mohammed Fadhil Ibrahim, Huda Khurshed Shawkat Aljader, Nawar Abbood Fadhil

Covid-19 pandemic affected our daily activities. Education is the fastest aspect that has been hindered due to the pandemic. The majority of the schools and universities were prone to global closure. The world turned toward using e-learning as an alternative to traditional learning due to the strict lockdown procedures. This study presents a model to measure the students’ intention to use e-learning during the pandemic in Iraq. Technology Acceptance Model (TAM) is utilized with some external variables. Task Technology Fit (TTF) is also adopted as a moderator for the proposed model. The model is implemented on a sample of students studying at Technical College of Management – Baghdad who have completed the academic year using e-learning. Eighteen hypotheses were formulated; out of them, only two were rejected. The model is tested and evaluated using PLS Tool. The results showed that the students have a positive intention to use e-learning during the covid-19 pandemic; also, most of the external variables were statistically significant. Furthermore, using TTF positively moderates the proposed model. The results open the way to perform more investigations in adopting e-learning during unusual situations.

517

Authors: Prabhakaran Mathialagan, Malathy Chidambaranathan

Upper Aero Digestive Tract (UADT) cancer is one of the most common cancer types in any gender. Early detection and diagnosing such type of cancer will reduce the risk of death in human. In this paper we addressed the state-of-the-art solutions for two major pathological constraints like, dysplasia type tumour grading and tumour grade classification with artifact’s present in biopsy images. A new Whole Slide and patch-based CNN model was proposed which involves different pre-processing techniques to detect and normalize the stain artifacts present in biopsy tissue samples. The proposed CNN model is developed to classify different oral cavity tumour sites and its gradings automatically. It was observed that the proposed CNN model achieved best accuracy of 98 % with zero classification error rate. These findings will throw a light for pathologists to handle digital biopsy images with artifacts and also will classify dysplasia type tumour grades automatically.

539

Authors: Dipak Patel, Kiran Amin

Nowadays, the rapid growth of internet has led to the way for easiest data generation. Sentiment Analysis (SA) is necessary to scrutinize the user-generated data for classifying the sentiment polarity into anyone of the two classifications (negative and positive). SA is implemented by machine learning techniques and lexicon oriented techniques. However, the traditional SA models could not deal with missing labelled data and internet slang data. This paper aims to propose a new multi-source sentiment analysis model that includes six stages to perform the task of domain adaptation. For checking similarity between statements, higher order statistics based on text features is exploited. The higher order statistics based features extraction includes use of modified cross entropy measure. Remaining features are then given to proposed classifier that predicts the polarity of target domain in a precise way. For classification purpose Neural Network (NN) is exploited. Particularly, the weights of NN are tuned in an optimal manner using Improved Grey Wolf Optimization (IGWO) algorithm, which is the enhanced version of GWO algorithm. Finally, the performance of the proposed model is compared over several conventional methods of domain adaptation and gained an improvement of 28% to 6% in accuracy.

559

Authors: Mohammed Amine Hassoine, Fouad Lahlou, Adnane Addaim, Abdessalam Ait Madi

This paper investigates the optimal layout of wind turbines (WTs) within a wind farm. Finding the best placement of WTs in a wind farm is a challenging process due to the existence of multiple wake effects. A biogeography based optimization (BBO) algorithm method is proposed to search for the optimal location of WTs in a wind farm (WF), to maximize the power produced by the WF and improve the annual economic performance of the WF. A wind turbine (WT) that operates in the wake of one or more other turbines is subject to lower flow and therefore produces less power. When designing a wind project, the arrangement of the turbines with each other in a wind farm is a very important factor. The best layout of a wind farm is to achieve the optimal placement of the turbines in relation to each other in a given area to maximize the efficiency of the whole wind farm and reduce its cost. A dense configuration would result in considerable power losses. Each turbine must have a sufficient distance from other turbines in the WF where the optimal number of turbines should be placed. The BBO approach is conducted on a 2 km x 2 km wind farm assuming a constant wind speed of 12 m/s with a fixed wind direction, for solving the wind farm layout optimization problem in two different configurations which include 26 and 30 WTs respectively. A comparison of the results obtained with the previous studies shows that the BBO is more efficient in terms of maximizing power output and economic profitability of the same wind farm model, which validates that BBO performs effectively in optimizing WTs placement within WF. BOO provides the greatest improvement in the optimal layout, for example, in the case of the layout for 30 WTs. The power output reaches 15,383 KW, the agreement between the ideal and the optimal layout is more favorable. The difference in output power is only 169 KW (1%). Knowing that the ideal layout is reached if all WTs receive a wind flow with a maximum wind speed of 12 m/s. Furthermore, a case study of wind turbine layout optimization using the BBO program on the Alta X wind farm has been performed under variable wind speed and variable wind direction. The results indicate that the optimized layout of the Alta X wind farm achieves a 12% increase in the power output for a similar cost when compared to the original layout of Alta X. It is also more appropriate for evaluating the wind farm layout in the Wind Power Project (WPP).

583

Authors: Abdelghani Aib, Djalal Eddine Khodja, Loutfi Benyettou, Salim Chakroune

This article provides a hardware implementation of torque and flux estimators for direct torque control (DTC) of an induction motor (IM) on the Field Programmable Gate Array (FPGA). Due to its high processing frequency, the FPGA circuit presents an alternative strategy for achieving a high performance DTC implementation. This cannot be achieved by Digital Signal Processing (DSP) or Microcontroller application. Optimized hardware architecture is also proposed for implementation of a complete estimator with explicit determination of all the quantities such as the flux phase, the flux amplitude and the torque amplitude. This work presents a precise hardware estimator of the torque and the flux, it developed with negligible estimation error (0.0002 [Wb] RMS estimate error for the stator flux and 0.0005 [Nm] RMS estimate error the torque), in addition to that this estimator is implemented on the Virtex-4 FPGA board with minimal resources(less 5% of slices registers, 5% of DSP48Es and around 17% of LUTS). This estimator is developed and synthesized independently of the command and can be reused for any task which requires the estimation of these quantities. After this the estimator was used to complete a DTC control in a hardware co-simulation procedure with the Xilinx system generator and MATLAB/SIMULINK tools on the Xilinx ML402 development kit.