International Journal of Intelligent Engineering and Systems

12

Authors: Mahesh Boddu, Soumitra Kumar Mandal

This article presents an efficient hardware architecture of lossless image compression mechanism using colour filter array (CFA) for wireless camera networks. Recently, Quantum-dot cellular automata (QCA) technology has been widely using in the field of VLSI, thus the proposed compressor is developed using QCA technology to mitigate the area, and power as compared to standard CMOS technology. In addition, an improved and extended Golomb-Rice entropy coder (IEGREC) is proposed to reduce the memory requirement, and computational complexity. The IEGREC approach eliminates the requirement of context module, which is a major component in lossless image compressor architectures and its memory to lessen the hardware resource utilization. Further, to maintain the pixel connectivity and higher compression ratio (CR), the proposed architecture utilizes the adaptive Golomb-Rice (GR) parameter prediction and control module. Finally, packer module is used to maintain the flow of output compressed bit stream. The extensive simulation results show that proposed QCA-IEGREC method performs superior as compared to state-of-art approaches in terms of lesser area as 3700 gate count, and reduced power of 1.02 mW, and even image quality statistics such as 18.78dB of peak signal-to-noise ratio (PSNR), 152.3 of figure-of-merit (FOM), and 29.3 of CR.

36

Authors: Omar Abdullah Murshed Farhan Alnaggar, Basavaraj Ningappa Jagadale, Swaroopa Hebbar Narayan

Early detection of the brain tumors can help in providing sufficient treatment to completely cure the disease. Machine Learning (ML) and Deep Learning (DL) algorithms are utilized predominantly in recent years for the tumor detection process from the brain Magnetic Resonance Imaging (MRI). In this paper, an effective tumor detection framework is proposed by utilizing a hybrid segmentation algorithm and a new hyper-parameter optimized Convolutional Neural Networks (CNN) based classifier using Chimp Optimization Algorithm (COA). The framework includes pre-processing, segmentation, feature extraction and classification stages. The pre-processing stage performs the denoising using wavelet filter and image enhancement using adaptive histogram equalization. Then the Boosted Crossbred Random Forests (BCRF) is used to segment the regions of interest to identify skull, tumor and lesions. The features are extracted using the Grey-Level Co-occurrence Matrix (GLCM) and Gabor filter. Finally, the tumor classification is achieved using the COA-CNN. The COA-CNN is developed by utilizing the COA to tune the hyper-parameters of CNN to obtain optimal CNN architecture. Experimental evaluation was performed on a dataset of 3264 MRI brain images containing normal and tumor images. No tumor, Glioma, Meningioma, and Pituitary tumor are classes of the MRI images which are correctly classified by the proposed BCRF and COA-CNN based detection model with a testing accuracy of 95.18% and reduced processing time. The results indicate the efficacy of the proposed method in classifying brain tumors for assisting physicians in early tumor diagnosis and treatments.

57

Authors: Fikri Budiman, Pulung Nurtantio Andono, De Rosal Ignatius Moses Setiadi

This research proposes an encryption method for images based on dual chaos systems and dual hash functions. SHA-256 and SHA-512 are hash functions chosen to process plain images and private keys. The purpose of using hash functions on plain and key images is to increase encryption resistance against differential attacks and perform local encryption in each image zone. The proposed method consists of two stages of encryption, the first involves dual hash functions and confusion with a chaotic system by performing encryption based on zoning and rotation of the image. The second stage is a diffuse process using a logistic map. The proposed method is proven to be strong against various attacks that have been tested with various measuring tools such as entropy, histogram analysis, chi-square, number of changing pixel rate (NPCR), unified averaged changed intensity (UACI), and the avalanche effect (AE). This method also proved to work better than the state-of-the-art method in this research, significantly based on entropy, UACI, and NPCR.

79

Authors: Nataraj Kalaivani, Raman Beena

Heterogeneous Cross-Project Defect Prediction (HCPDP) predicts the defects by incorporating the project datasets with different distribution and metrics. However, the class imbalance problem and shallow feature learning of classifiers reduce the overall performance. This paper presents an HCPDP using Improved Synthetic Minority Oversampling Technique (ISMOTE) and hybrid deep learning classifier of Golden Eagle Optimized Siamese Neural Networks (GEO-SNN). Initially, Peters filter removes irrelevant instances in source-target dataset based on heterogeneous metrics. ISMOTE, enhanced through fuzzy mutual information based KNN attribute weights assignment, is applied solve the class imbalance problem. Features are scaled using Z-Score standardization and metric matching by Spearman's Rank Correlation (SRC). Finally, GEO-SNN, developed by SNN parameter optimization using GEO, performs deep semantic feature learning for defect prediction. Experiments using benchmark datasets showed that the proposed ISMOTE and GEO-SNN based HCPDP technique has provided 99% defect prediction accuracy and reduced processing time by 20%than the state-of-the-art methods.

99

Authors: Sudhakar Karna Narayana, Naveen Thimmahanumaiah Hosur

Wireless Mesh Networks (WMNs) are considered as an important structure for the next generation wireless networks that have numerous nodes organized in the mesh topology. In WMN, the data packets are forwarded through the multi-hop model. Nowadays, the combination of mesh clients with the Internet of Things (IoT) is gaining substantial importance for connecting numerous machines and it is used to acquire faster coverage. Moreover, the nodes in the WMN are susceptible to the malicious nodes which transmit the compromised and manipulated data to the destination. In this research, the Priority based Trust Efficient Routing using Ant Colony Optimization (PTER-ACO) is proposed to achieve data privacy and reliable routing over the WMN. Subsequently, Advanced Encryption Standard (AES) and Rivest Cipher 6 (RC6) are used to provide added security to the data. The PTER-ACO is analyzed by means of energy efficiency, network throughput, Packet Loss Rate (PLR), computational overhead, and latency. The existing methods such as Improved version of the Dijkstra Algorithm (IDA), Round Trip Time (RTT) based detection and Robust and Trusted Scheme (RTS) are used to evaluate the PTER-ACO, against these methods. The PLR of the PTER-ACO method is 21 % for 10 malicious nodes, which is less when compared to the IDA, RTT and RTS.

116

Authors: Madha Bramarambika

The MRI is assisted by doctors to evaluate the tumors that help them for planning the required treatment. The physicians are dependent on the medical MRI diagnosis but that required lots of effort in finding out the tumor regions. The existing models speed up the scanning process using the MRI technique that reduced the burdens from radiologists but showed abnormality due to the presence of outliers when extracting hyperparameters. The proposed Hybrid GLCM-LDTP-Le-Net Feature extraction technique is developed by combining the Grey level covariance Matrix (GLCM), Local Direction Ternary Pattern (LDTP) and Le-Net features. The feature extraction technique results with hybrid parameters, which are applied to the Bidirectional Long Short Term Memory (Bi-LSTM) model to classify the brain MRI as a Malignant or Benign. The proposed hybrid feature extraction increases the accuracy of learned models by extracting the combined features from the input data. The proposed hybrid feature extraction obtained an accuracy of 97.88 %, better when compared with the existing CNN-based models' methods that obtained accuracy of 91%.

135

Authors: Agus Subhan Akbar, Chastine Fatichah, Nanik Suciati

Many studies have been carried out to segmentation brain tumors on 3D Magnetic Resonance Imaging (MRI) images with 3D or 2D approaches. The 3D approach pays attention to the interrelationships between slices in a 3D image. However, this requires high resources, while the 2D approach requires lower resources but ignores the voxel relationship in 3D space. The 2.5D approach seeks to combine the lightness of the 2D approach and the voxel interconnection of the 3D approach. This article proposes SDA-UNet2.5D, a shallow UNet 2.5D architecture that pays attention to the interconnectedness of 3D images by involving five slices to get one slice of segmentation prediction results. The architecture is trained using the Brain Tumor Segmentation (BraTS) 2018, 2019, and 2020 datasets. Compared to other architectures, this proposed architecture has a high segmentation speed with 4.05-4.24 seconds to segment one patient data. Online validation resulted in superior average dice performance of 75.70, 88.82, 77.33 for the BraTS 2018, 71.29, 88.00, 76.55 for the BraTS 2019, and 70.80, 87.95, 75.89 for the BraTS 2020 validation datasets in the areas of Enhanced Tumor (ET), Whole Tumor (WT), and Tumor Core (TC).

163

Authors: Arun Prasath Govindan, Annapurani Kumarappan

Hand gestures and voice inputs have been measured as the vital communication component for the past few decades. Here, deep learning-based Reversible Convolutional Neural Network (Rev-CNN) is explicitly modelled to predict gesture-based sign language. Similarly, this work concentrates on modelling a reversible model to identify the voice gesture into sign language. It shows reversible representation by attaining superior accuracy with a lesser amount of model parameters and various CNN architecture. Here, the efficiency of the reversible model is evaluated with the prevailing G-CNN, VGG-11/16 model over the testing and training environment. Here, two diverse datasets like ROBITA Indian Sign Language Gesture Database and the standard voice-input dataset, is considered for evaluation purpose. The highest prediction accuracy of 94.38 % and 97.89 % is attained using the proposed reversible CNN model over the other approaches like GCNN, VGG-11 and VGG-16 model. The experimental outcomes and metrics like loss function, error rate and execution time are measured and compared with different methods like GCNN, VGG-11/16. Additionally, other efficiency metrics are utilized to determine the efficiency of the anticipated model. The model outperforms the existing approaches by categorizing the gestures with reduced error rate. The prediction accuracy of the reversible CNN (dataset 1) is 95.38 % and for dataset 2 is 96.69%. Similarly, the execution time is 5.5 minutes.

190

Authors: Santhosh Kumar Gorva, Latha Channagiri Anandachar

In recent decades, the rapid growth of cloud computing has led to increasing concerns about the security and energy requirement of cloud data centers. To overcome the concerns of load balancing and data security, a new hybrid model is proposed in this article. Firstly, a Modified Particle Swarm Optimization (MPSO) technique is proposed for balancing the tasks between heavy-loaded Virtual Machine (VMs) and light-loaded VMs. An effective load balancing improves the performance and availability of website, database, application and other computing resources. In the MPSO, a linear decreasing inertia weight is included to achieve optimal solutions in both VM migrations and load balancing. Secondly, an Enhanced Elliptic Curve Cryptography (EECC) algorithm is proposed for effective data security. In the proposed EECC algorithm, a new pseudo-random key is combined with a public key to improve the security of cloud data centers. In this work, the effectiveness of the proposed MPSO-EECC model is validated in terms of Service Level Agreement (SLA), execution time, energy consumption, energy SLA violation, encryption, and decryption comparison time. As represented in the experimental section, the proposed MPSO technique almost reduced 30%-70% of energy consumption compared to the existing optimization techniques. Similarly, the proposed EECC algorithm reduced 10ms to 30ms of decryption comparison time related to the comparative cryptography algorithms.

213

Authors: Kawther Dawood Salman, Ekhlas Kadhum Hamza

Due to the rapid growth in mobile data traffic, academia and industry are considering Li-Fi (Light Fidelity) as an alternative to Wi-Fi (Wireless Fidelity). Li-Fi is a high-speed, two-way wireless network system that has several advantages over traditional radio frequency systems. To fully exploit the potential of Li-Fi systems, users must be provided with a seamless connection, which must be achieved by carefully studying user mobility. Nowadays, most wireless Li-Fi network applications focus their design on solving some unspecific user-location problems that may fully achieve some preferred results in the domain they serve. For example, locating an unknown user in an indoor environment is considered a kind of solution for maintaining the QoS (Quality of Service) and delivering the signal to the user without distraction. Where the main research problem lies in the signal interference, loss, and scattering in the Li-Fi network, it is a big problem because it reduces the network performance and coverage. Therefore, to solve this problem, this research investigates the locating of the mobile user within the Li-Fi equipped room to obtain the best possible coverage of the user's service. The aim of the research was divided into two stages. In the first stage, the method used in our work was presented by showing the triangulation method in simulating user location data (sensor simulation), the first stage was not affected by noise, as it serves as proof of the correctness of our work and depends in the second stage when it is under the influence the noise. In the second stage, the method of RSS - triangulation (Receiving Signal Strength -Triangulation) was adopted to simulate the user’s location data (a simulation of the sensor) and under the influence of noise and implemented on a deep learning algorithm to determine the accuracy of the user’s location and how to deal with noise. In the simulation results for the data in the first stage, our work was validated using the triangulation method in data simulation. The average error for the X-Axis was 2.17344 〖× 10〗^(-14) cm; Y-Axis was 6.44762 〖× 10〗^(-14) cm and Z-Axis 4.65690 〖× 10〗^(-11) cm the results obtained from the triangulation simulation were close to the real results. In the second stage, when adopting the RSS-Triangulation method under effect the noise, the average error was achieved on the X-axis 2.05612 〖×10〗^(-3 )cm, Y-Axis was 4.56249〖 ×10〗^(-3 )cm, and Z-Axis 5.10474 〖×10〗^(-3 )cm where The DNN (Deep Neural Network) showed how to deal with noise, the highest error was obtained on the X-axis 2.51535 cm, the Y-axis 2.25903 cm, and the Z-axis 4.22898 cm for an indoor environment of 5m x 5m x 3m.

243

Authors: Nirmala John, Varaprasad Janamala, Joseph Rodrigues

Due to numerous operational restrictions and economic purposes, optimal load management for energy balance in the smart grid (SG) is one of the compensating responsibilities. This research provides a novel multi-objective optimization technique for attaining energy balance in SG, with the goal of avoiding fines due to excessive upstream network power extraction beyond contractual demand. Due to a lack of capacity to create the whole optimization towards the global optimum after each run, optimal load control (OLC) is a prevalent challenge. Adaptive-TLBO, the most recent variation of Teaching Learning Based Optimization (TLBO), comprises both alterations during the exploitation and exploration phases (ATLBO). Because the ATLBO is used on a modified IEEE 33-bus system, the results obtained in this mode are extraordinary. The energy balance has improved in addition to the enhancement of the voltage profile and the reduction of distribution losses. As evidenced by comparisons with PSO, basic TLBO, backtracking search algorithm (BSA), and cuckoo search algorithms, the suggested ATLBO algorithm has precedence over any other proposed algorithm (CSA).

260

Authors: Anthony Anggrawan, Mayadi, Christofer Satria, Lalu Ganda Rady Putra

Providing scholarships to students is an important action in helping students succeed in their studies. The scholarship recipients are students who meet the specified criteria. However, it is not easy to decide which scholarship recipients are genuinely eligible to receive scholarships, especially if the selection process for prospective scholarship recipients is done manually. Many prospective scholarship recipients must be selected, and more than one criterion is required. That is why it is not surprising that awarding scholarships to selected candidates takes a long time and the accuracy of the results is dubious. It means there needs to be a system that can assist in choosing prospective students who meet the criteria for the best requirements as scholarship recipients. Therefore, the purpose of this research is to build an application system that can provide recommendations for eligible candidates for scholarship recipients based on consideration of the criteria possessed by prospective scholarship recipients. This research method is an experimental study using a combination of the Analytical Hierarchy Process (AHP) and the Multi-Objective Optimization Method by Ratio Analysis (Moora) method to rank scholarship recipients. The study results indicate that using an application system integrated into the database in recommending prospective scholarship recipients according to the desired conditions is more accurate using a combination of AHP and Moora methods (up to 94.07%) than using only one Moora method (which is only 89.62%). This means that the selection of scholarship recipients according to the desired qualifications using the AHP method for weighting and the Moora method for ranking is a combination of methods that have relatively more reliable accuracy.

287

Authors: Salwa Khalid Abdulateef

The image vectorization based on geometrical contour representation is an active research topic. It involves approximating the contour of the objects inside the image using geometrical definition or estimation. Typical approach of estimating the parameters of the geometrical model is fitting-based estimation. However, the noise that exists in the image causes degradation in the performance. In order to overcome this degradation, we propose integrating evolutionary based optimization. In this article, the design of the operators of the genetic optimization for improving the contour detection results of segments-based estimation of the contour is proposed. The operator includes both crossover and mutation. The evaluation shows that Structural Similarity Index Metrics (SSIM) measure has increased after applying it from 0.64 to 0.77 on UEC Food 100 Dataset and from 0.54 to 0.66 after applying it to UEC Food 256 Dataset.

307

Authors: Mays D. AL-Bayati, Duraid Y. Mohammed, Mohammad Sarfraz

Arrhythmia identification using electrocardiograms (ECGs) is a critical component of biomedical signal processing and pattern recognition. The classification of ECG arrhythmias is presented in this article using features extracted from the Extreme Value distributor of the probability density function (PDF), also training model built using the Alexnet architecture of a conventional neural network (CNN). A continuous wavelet transform (CWT) was used to convert the features space to a two- dimensional image. Additionally, a comparison of the two approaches was conducted. In the first case, we used CWT to convert ECG data to two-dimensional images. In the second case, features were extracted from PDF files and converted to the image domain using a CWT. With a learning rate of 0.001 and batch size of 250, the accuracy of tests was 98.67 %. The results indicated that the proposed method is more accurate than the conventional method when it comes to feature extraction.

329

Authors: Korawitch Kaiyawong, Keerati Chayakulkheeree

This paper proposes a mixed-integer particle swarm optimization (MIPSO) for coordinated optimal placement of energy storage system (ESS) and capacitor bank (CB). In the propose method, optimal ESS scheduling (OESSS) is solved by particle swarm optimization (PSO), as a subproblem, the optimal coordinated placement (COP) for ESS and CB, simultaneously. The distribution system annual loss minimization (DSALM) is used as the objectives of COP problem. The proposed method was tested with the IEEE 33-bus radial distribution test system. The results demonstrated that the proposed method is successful and robust in minimizing system losses, which loss saving of 35.12% when compare to the based case, which is the best solution among other existing methods.

348

Authors: Hayder M. A. Ghanimi, Shaymah Akram Yasear

The global pandemic (Covid-2019) has severely affected all aspects of our life and even changed the way we live and work. As the pandemic outspread, healthcare professionals need urgent interventions to control the harmful consequences. Encountering such a crisis makes them more prone to negative psychological ramifications in their work environment, making them unable to provide the proper support. Burnout is the most negative feeling increased among healthcare professionals while compacting the virus. One essential move to remedy the impacts of burnout is understanding its determinants and their causal relationships. This paper addresses the design of a computational (cognitive) agent model of burnout for healthcare professionals using a temporal-causal network model. Several determinants of burnout with their causal relationships were identified from the literature and formalised to construct the proposed cognitive agent model. In addition, different simulation experiments were implemented to obtain a clear insight into the causal relationships among burnout determinants and those experiments are exhibited similar behaviours to exiting literature. Furthermore, the developed model was evaluated using two different methods: mathematical analysis to prove its implementation was achieved right and automated logical verification to check several properties as shown in the literature to confirm that the suitable model was built. The obtained cognitive agent model could be helpful to develop a covid-19-aware analytics software agent that can monitor healthcare professionals' mental health.

371

Authors: Tossapol Kiatcharoenpol, Sakon Klongboonjit

A reliable and sensitive technique for predicting quality of a plastic work-piece produced in injection molding process is essential help for practicing engineers. A system based on the process parameters that can estimate both two prime characteristics, %volume shrinkage and warpage of work-piece before it produced is significantly beneficial. In this paper, a fast feed forward network, Hybrid Neural Network (HNN), is proposed to construct the predictive model for those two quality characteristics. The unique algorithm of HNN based on the optimization of the weights of each layer is changed to a linear problem by linearization of the sigmoid functions. As iteration procedure used in Backpropagation algorithm is eliminated, the network training time is significant reduced. With this fast convergence of using HNN, the intelligent predictive model for injection molding process that can learn online is possible for further study. To entitle the network to cater for various process parameter conditions, a knowledge base as training and testing data have to be generated on the experimental data in a comprehensive working range of a plastic injection molding process. Consequently, the experiments were performed in 256 conditions based on the combination of nine basic process parameters. The neural networks were trained and the architecture of networks was appropriately selected by benchmarking the Root Mean Square error (RMS). The results of the novel network, HNN, have shown the ability to accurately predict the percentage of volume shrinkage with the 1.02% and 4.87% error at training and testing stages, respectively and for warpage with the 3.76% and 2.47% error at training and testing stages, respectively. These accuracy results are similar to those of backpropagation neural network (BPNN), but HNN has shown the superior fast converging about 38.5% and 66.7% over than those of BPNN.

394

Authors: Malikhah Malikhah, Riyanarto Sarno, Sozo Inoue, M. Syauqi Hanif Ardani, Doni Putra Purbawa, Shoffi Izza Sabilla, Kelly Rossa Sungkono, Chastine Fatichah, Dwi Sunaryono, Arief Bakhtiar, Libriansyah, Cita RS. Prakoeswa, Damayanti Tinduh, Yetti Hernaningsih

Viral respiratory infections are the most common diseases suffered by all age groups worldwide. The gold standard for diagnosing viral respiratory infection is through the molecular method, but this diagnosis is expensive, requires sophisticated equipment, can only be performed by well-trained medical staff, and is painful. Volatile Organic Compounds (VOCs) are compounds released from the human body that can be a marker of disease and based on numerous studies it also contains VOCs. An electronic nose (E-nose) is a device that can be used to identify disease. This study proposes a new approach for the detection of viral respiratory infections through sweat from the armpit using an E-nose consisting of 5 semiconductor gases and a single-board computer. Several statistical parameters are used to obtain features and the detection algorithm used is Fully Connected Deep Network (FCDN). Several sizes of hidden layers were tested to obtain the best FCDN model. This study also proposes the selection of the best FCDN model which is a trade-off between complexity and accuracy, so that the model stored in E-nose is a model that not only has good accuracy but is also not too complex. The experimental results show that using 29 statistical parameters and 2 hidden layers generate the highest accuracy of 0.940 for the detection of 2 classes, namely positive and negative, with sensitivity and specificity of 0.967 and 0.915, respectively, where the best FCDN model has a total of 90,561 parameters.

414

Authors: Ravindar Mogili, G. Narsimha

Abnormal behaviour of the heart called arrhythmia can be recorded as an electrocardiogram (ECG) signal. ECG plays a vital role in the diagnosis of heart disease. Advances in machine learning allow the development of computer-aided diagnostic models to identify heart disease type. We have proposed a hybrid model with Convolution Neural Network (CNN) to auto extract features from ECG and use XGBoost to assess the type of arrhythmia. We tested our model to diagnosis eleven kinds of arrhythmia beats from the MIT-BIH arrhythmia database and obtained overall sensitivity of 92.61%, specificity of 99.85%, positive predictive value of 95.99% and accuracy of 99.84%. The robustness of the proposed model is further confirmed by classifying the arrhythmia beats into 5 classes according to the AAMI standard and comparing the results with the state-of-the-art methods. Attained overall Sensitivity of 94.36%, specificity of 99.44%, positive predicative value of 96.40% and Accuracy of 99.69% for 5 AAMI Classes. The results demonstrated that the proposed model could be used in the diagnosis of arrhythmia.

441

Authors: Rahmad Hidayat, Agus Harjoko, Aina Musdholifah

An image retrieval system is required to provide high accuracy in a short time. Combining various features will usually increase accuracy but also increase retrieval time. This study developed a CBIR (Content-Based Image Retrieval) method based on hierarchical clustering on low-level features. Low-level features consisting of color, texture, and shape are extracted and then clustered hierarchically. The resulting clusters are then validated to obtain their optimal number. In the retrieval process, the query image features are extracted and compared with the cluster centroid on each feature. The scores of query results on each feature are normalized, and then the normalized scores are weighted to get the total score. The experiment was carried out using three datasets, namely DIKE20, Corel-1k, and Corel-10k. Based on the experimental result, the proposed method shows better performance compared to the existing state-of-the-art method. On the Corel-1k and Corel-10k datasets, the proposed method obtained precision scores of 0.81 and 0.62, respectively.

464

Authors: Aicha El Filali, El Habib Ben Lahmer, Sanaa El Filali, Mohammed Kasbouya, Mohammed Amine Ajouary, Saadia Akantous

The covid 19 has caused a strong health and economic crisis, where the business environment is turning into a great uncertainty and customer demand has become more and more fluctuating. As a result, demand forecasting as well as the task of predictive analysis, holds a great attention in the supply chain, in order to meet the needs of customers, avoid any out of stock, and at the same time avoid a waste of resources. In this paper, we propose a deep learning method based on long-term memory multilayer networks (LSTM) for demand forecasting. Using the grid search method, our method has the ability to automatically select the most optimal forecasting model, considering different combinations of LSTM hyperparameters of a time series. The proposed model has shown strength in capturing the existing nonlinear features in time series data compared to some known time series forecasting methods derived from statistical and machine learning approaches, using historical sales data of a Moroccan pharmaceutical manufacturing company. These methods include exponential smoothing (ETS), autoregressive integrated moving average (ARIMA), recurrent neural network (RNN). The evaluation of the proposed method and the comparison methods was performed using the root mean square error (RMSE) and Symmetric mean absolute percentage error (SMAPE). The comparison of the test results showed that the proposed method is the best performing with RMSE of 4487.32 and SMAPE of 0.026 much better than those obtained by the other models.

494

Authors: Asha Kethaganahalli Hanumanthaiah, Manjunathswamy Byranahalli Eraiah

Multimodal biometric systems are a natural evolution of traditional biometric authentication systems with use of multiple biometric to enhance the security level. Presentation attacks in form of photos and replay videos are becoming a threat to the biometric authentication systems and it becomes necessary to prevent from these attacks for an enhanced security level. This work proposes a challenge response based multi modal biometric authentication resilient to presentation attack. The proposed system is designed for multi modal biometric involving face and iris biometrics with challenge in form of arbitrary emotion invoking image at a random position on the screen. The changes in region of interest around the Facial landmarks and distance of iris to Facial landmarks are captured as response. The response is then matched with expected response derived based on the position of arbitrary emotion invoking image on screen to detect presentation attacks. In addition the work also proposes a hybrid deep feature which provides an addition security in recognition process to ensure authentication failure for fake samples. The error rate in leakage of spoofs is very low in proposed solution (less than 1%) in different environments compared to existing works. The recognition accuracy of face is atleast 1% higher and atleast 2% higher for IRIS in proposed solution compared to existing works.

521

Authors: Arjun Singh Saud, Subarna Shakya

Although some studies employed technical indicators as input to machine learning models to forecast stock trading signals, there was a noticeable gap between how technical analysts and machine learning professionals used technical indicators. Based on this finding, this study suggested a know sure thing based machine learning (KST-ML) technique for anticipating stock trading signals and compared its results to KST-based trading and Buy-Hold strategies. The main idea behind the KST-ML strategy is to learn and then predict stock trading signals from the relationship between KST indicators such that more accurate predictions can be made. The strategies were assessed based on their annual rate of return (ARR), Sharpe ratio (SR), and percentage of profitable trades executed. In addition, the proposed strategy's performance was compared to that of several intelligent stock trading strategies proposed in the literature. The proposed method clearly outperformed the other two strategies in terms of all three evaluation measures. With a 5.6 SR value, the strategy generated a 67.73% ARR and made 79.38% profitable trades. The results were much better than the results from the KST-based and the Buy-Hold strategies. Furthermore, we found that the annual returns generated by the KST-ML approach were significantly higher than those generated by other intelligent techniques reported in the literature.

543

Authors: Ahmed H. Mohammed, Ahmed Kareem Shibeeb, Mohammed Hussein Ahmed

Recently, the proliferation of the Internet of Things (IoT) services and applications has prompted several governments to deploy many cameras in critical sites, such as embassies, military institutions, and ministries. The images must be encrypted to protect data shared between IoT sensors/devices and embedded subsystems. Often, IoT devices are smaller and less powerful. A lot of traditional encryption methods are computationally inefficient. They require many rounds of coding, which takes up much space on a device. Despite this, this paper has developed a lightweight encryption algorithm based on chaos theory suitable for IoT devices with limited processing power. The confusion-diffusion structure is often divided into two independent components in chaotic image cryptosystems. However, it reduces the security of the cryptosystem since the independent design may be subjected to cryptanalysis individually. This paper presents an efficient chaotic image encryption algorithm based on a plaintext-associated approach. It incorporates confusion-diffusion operation to increase encryption reliability. The Zaslavsky map initial values are determined by the values and locations of an input image and random values to resist chosen-plaintext cryptanalysis. Furthermore, the proposed method uses a simultaneous confusion-diffusion process to resist any separate attack. The simulation and experimental analysis demonstrate that the proposed encryption algorithm has 0.3927 seconds as a fast encryption time. Moreover, it provides a high performance compared to several other chaotic-based image cryptosystems. as measured by the ‎histogram, entropy (‎7.9972‎‎3), ‎ pixel change rate (‎99.7194‎), unified average changing intensity (‎33.4635‎) within the critical interval, pixel ‎correlation, global and local entropy, CDR (ciphertext difference rate), and gray difference degree (GDD). It also has ‎higher security than other chaotic-based image cryptosystems.

563

Authors: Kiattisin Kanjanawanishkul

Eri silkworm pupae are well known as an alternative for a protein food source. At present, they are sold as canned food for long-term preservation. Therefore, good quality and size consistency are essential. To evaluate quality and size, an image-based grading method was proposed. The image of pupae was captured and then shape features (i.e., solidity, aspect ratio, and extent) and color features based on three color models (i.e., RGB, HSV, and L*a*b*) were extracted. Two neural networks with 10-fold cross validation were separately developed for shape evaluation and color evaluation. After misshapen and discolored pupae were identified by neural networks, remaining pupae were graded into five size numbers according to their length: very small, small, medium, large, very large. Experimental results showed that the average accuracies for shape evaluation and color evaluation were 99.64% and 99.58%, respectively. The accuracy for size evaluation was 94%. Therefore, the proposed grading method reduces sorting time and increases sorting accuracy.

584

Authors: Umar Umar, Faanzir Faanzir, Firdaus Firdaus, Indri Suryawati, Muhira Dzar Faraby, Adi Soeprijanto, Ontoseno Penangsang

The interconnection of single-phase households DG into the existing grid requires appropriate anticipation to ensure the quality and reliability of the existing system. This paper deals with the optimal placement and sizing of multi single-phase DG in unbalanced distribution systems for power loss reduction and voltage improvement while maintaining voltage unbalance and harmonic are in the acceptable limit. Optimal location and sizing of DG are determined simultaneously using the improvement version of Symbiotic Organisms Search (SOS), called Dual-Phase Parasitism SOS (DPP-SOS) with a crossover operator. DPP-SOS consists of original parasitism (OP) and random weight parasitism (RWP). DPP-SOS was programmed using MATLAB software and validated using 26 benchmark functions. To demonstrate the effectiveness of this method in reducing losses, the DPP-SOS is also tested using 7, and 25 bus test systems. Result show that DPP-SOS with a crossover operator has been successful in determining the location and rating of single-phase and three-phases DG simultaneously. DPP-SOS has an average convergence speed of 24.08% faster than the basic SOS. The simulation result also shows that multi single-phase DG is effective in reducing power loss, increasing voltage, and reducing voltage unbalance in the distribution system.