Cyber Security and Computer Science

This paper presents the framework for the optimized development of a digital platform based on ISO/IEC-27001 with the objective of making an initial diagnosis regarding the informatics security level in any company. In addition, the optimization process considers that the diagnostic results should be clear and direct, to making possible the fast security risk mitigation. In particular, the optimization process is based on the analysis of a conventional Management Information System framework in order to propose a novel customized framework for ISO/IEC-27001 applications. Thus, an optimized Management Information System is proposed which is the basis of the optimized digital platform. As preliminary results, the reduction of needed elements for the initial diagnosis for the informatics security promotes the simplicity of the application and thus, increases the possibility of applying the ISO/IEC-27001 to a greater amount of users, which means that it is promoted cybersecurity.

Intrusion Detection System (IDS) is used to protect a system or a computer network from different kinds of anomaly attacks. Different detection techniques have been discussed on network-based IDS. The study has been done on the operational procedures of network based open source IDS tool Snort based intrusion detection system, which can read every incoming or outgoing packet through a network and alert the admin accordingly. In this paper, Different types of IDS are compared and criticized which explores the vulnerability of the system. To check every packet, Snort uses a central database system of signature. A layered database system has been proposed to upgrade system performance. An analytical operation has been conveyed on the proposed solution and compared with the existing standard system. After applying the proposed solution the number of packets analyzed rate has been increasing remarkably from 86% to 98%.

With the rapid development of technology, one of the most important requirements in today’s systems is the reliable transfer of information and confidentiality. Thus, military, electronics, banking systems and many other places have become the fields of use of cryptography science. Cryptology methods are used to solve these problems and need of secure information transfer resulted in the development of reliable encryption techniques. In this study, a new poly-alphabetic substitution cipher is designed using the coordinate axes. The proposed method of a hybrid encryption method is a mix of the Polybius square cipher and the Vigenère cipher, reinforced with the RSA cryptography algorithm. For each letter in the alphabet, there is more than one point on Cartesian coordinate system and in the calculation of these points random chosen values are used. Values used to calculate alphabet and start index of alphabet are send to receiver using RSA algorithm with cipher text. So, with multiple security stages the proposed method is a strong encryption method that is difficult to decode.

In the construction of the wireless sensor network; MAC protocols are considered as a foremost element of IoT based devices that are used for data transmission and greater scalability with simple executions. Furthermore, most of the researches on WSN conduct directional antenna to provide substantial progresses in communication. However, numerous types of problems such as deafness, deadlock, and head of line blocking problems irrespective of a number of channels are introducing inherently in directional MAC protocols. Some proposed asynchronous multichannel MAC protocol is based on RIT or RI mechanisms that escape those prevalent problems. Using a multichannel directional MAC with RP is proposed in this article that referred to as Receiver Pivotal DMAC (RP-DMAC). We propose RP-DMAC as a state-of-the-art protocol for IoT based WSN to separate deafness and other problem by using directional RTR (Ready to Receive) packet, data channel, and guard band. As like most of the DMAC protocols, in default mode, our RP-DMAC is sender-initiated. When deafness problem is recognized then as our RP protocol manner; nodes will initiate with DRTR negotiation in its unused sector expending through guard band to received data from desire sender. To solve the deadlock and packet loss, our proposed RP protocol performs better results than DA-MAC and circular RTR MAC in terms of packet drop ratio, overhead, and throughput.

The increased usage and popularity of Android devices encourage malware developers to generate newer ways to launch malware in different packaged forms in different applications. These malware causes various information leakage and money lost. For example, only in Canada, McAfee, which surveyed 1,000 Canadians and found 65% of them, had lost more than $100 and almost a third had lost more than $500 to various cyber scams so far this year. Moreover, after identifying software as malware, unethical developer repackages the detected one and again launches the software. Unfortunately, repackaged software remains undetected mostly. In this research three different tasks were done. Comparing to the existing work we have used source code based analysis using bag-of words algorithm in machine learning. By modifying Bag-of-word procedure and adding some additional preprocessing of dataset the evaluation results represent 0.55% better than the existing work in this field. In that case re-packaging was included and this is a new edition in this field of research. Moreover in this research, a vocabulary was also created to identify the malicious code. Here with existing 69 malicious patterns more 12 malicious patterns were added. In addition to these two contributions, we have also implemented our model in a web application to test. This paper represents such a model, which will help the developers or antivirus launcher to detect malware if it is repackaged. This vocabulary will also help to do so.

Observing and experimenting with malware with full user control have been complex and difficult to say the least. As time goes on, malwares are becoming more advanced and has the ability to realize that the environment they are targeting is virtual, thus shutting their process and leaves the testers unable to analyze further. To combat this problem, a sandbox can be used to test these malwares through modifications. The sandbox is needed to create a dummy virtual environment to test the malwares on, and modifications on the said environment will allow more controlled and specified testing. Bypassing intelligent Malware for in depth analysis will be successful. Dynamic analysis will be performed, specifically agent-based using Cuckoo open-source sandbox and agent-less using DRAKVUF by hypervisor and virtualization extension. Analysis result will be classified over few pre-defined criteria including network requests, system injections and modifications, security measures and kernel alteration; ultimately proving which technique is appropriate and reliable for prominent malware analysis.

This paper represents a simulation-based investigation of permissions in obfuscated android malware. Android malware detection has become a challenging and emerging area to research in information security because of the rapid growth of android based smartphone users. To detect malwares in android, permissions to access the functionality of android devices play an important role. Researchers now can easily detect the android malwares whose patterns have already been identified. However, recently attackers started to use obfuscation techniques to make the malwares unintelligible. For that reason, it’s necessary to identify the pattern used by attackers to obfuscate the malwares. In this paper, a large-scale investigation has been performed by developing python scripts to extract the pattern of permissions from an obfuscated malwares dataset named Android PRAGuard Dataset. Finally, the patterns in a matrix form has been found and stored in a Comma Separated Values (CSV) file which will lead to the fundamental basis of detecting the obfuscated malwares.

Pixel selection for data hiding becomes crucial for the solutions in spatial domain of steganography to ensure imperceptibility. This paper presents an efficient approach of pixel selection technique for hiding secret data in cover object of image steganography. After reviewing recent literature, it has been observed that most of the works on pixel selection uses zig-zag technique for their solution. However, it becomes very prone to steganalysis based attacks by the intruders. In this study, 8-directions pixel selection technique is proposed to embed data in the cover image where Least Significant Bit (LSB) method has been used on Red, Green and Blue (RGB) color image especially focused on JPG, JPEG, and PNG. Since this projected procedure avoids the known steganalysis techniques, it will be challenging for the attacker to recognize the presence of secret information from the stego image. To measure the quality, statistical analysis has been performed where the value of the quality measurement matrices has provided better results.

Kerberos, a renowned token based authentication protocol, which is famous since mid-80’s for its cryptographic process, assurance of privacy, and data security for identifying appropriate users. Due to its versatile characteristics, users of the system often need to remember complex passwords as the good practice of the method requires update of the same within a defined time-frame which becomes bit difficult for users to cope up with. At the same time, it also not provides adequate channel security to transmit the user credential between the pathway of the client and server. Therefore, researchers are trying to find out a simple solution where user does not necessitate to memorize the passwords where it could guarantee better user validation. In this paper, an enhancement of Kerberos authentication model has been proposed where biometric template and Steganography are incorporated to solve the existing weaknesses. Instead of taking username and password, the new solution will take a pair of random fingerprints from the user and convert it into a hash. It will then embed the hash in the randomized image and send it to the server for authentication. A security analysis of the proposed protocol is proven using BAN logic in this article where it ensures reliability, practicability and security of the enhanced Kerberos protocol.

In the era of technology, attack on computer infrastructure is considered as the most severe threat. Web server is one of the most important components of this infrastructure. Preventive measures must be taken to deal with these attacks on the web servers. For this reason, vulnerability detection needs to be carried out in an effective way and should be mitigated as soon as possible. In this paper, an effective framework for vulnerability detection of web application is proposed. This framework targets the web applications developed with content management systems (CMSs). It obtains prior knowledge of the vulnerable extensions of a specific CMS from its contributors. The framework is run against a target web server using a well-known port scanning tool, Nmap. It checks if there is any existing matches for the vulnerable extension installed in that web application. Finally, the framework gives an output comprised of the installed extensions along with the installed vulnerable extensions in that web application. Although the output result is shown in the Nmap console, the framework is a segregated entity that works in collaboration with Nmap. Thus this framework can be well-utilized by the security specialists to assess the security of a web application in an easier and effective way and also to evaluate vulnerability of web servers; hence shielding the web applications from various kinds of security threats.

E-commerce plays a significant role to grow its business globally by satisfying the modern consumer’s expectations. Without the help of Operating System (OS), e-commerce applications cannot be operated as well as broadcasted on the web. It is evident after analyzing this study that web administrators of the business are sometimes being careless, in some cases unaware about the risk of cyber-attack due the lack of vulnerability research on their OS. Therefore, a good number of the e-commerce applications are faced different type of OS exploitations through different types of attack e.g. denial of service, bypass, DECOVF, etc. that breaches the OS’s confidentiality, integrity and availability. In this paper, we analyzed 140 e-commerce sites servers’ information and its related 1138 vulnerabilities information to examine the risks and risky versions of the OS in e-commerce business. The probabilities of vulnerability are calculated using Orange 3 and feature selection operation has been performed using Weka through IBM statistical tool SPSS. This study identifies few versions of Ubuntu that are found in critical status in terms of risk position.

Since twenty years, several t-way strategies have been developed for Combinatorial Input Interaction (CII) based system to reduce the number of test cases in the test suite. The t-way strategy can be applied to Sequence-less CII system, where all inputs are parameterized and parallel. From the literature, the searching methods used in t-way strategies are divided into deterministic and non-deterministic for reducing test cases for all test configurations. It is found that t-way strategy is an NP-hard problem; no deterministic and non-deterministic t-way strategies claim that their strategy can generate the optimal number of test cases for all test configurations. In this research, an Interactive t-way Test Case Generation (ITCG) algorithm is proposed inspiring fish swarm searching algorithm to integrate with t-way strategy and evaluate the generated number of test cases comparing with existing renown t-way strategies. The results show that the proposed t-way test case generation inspiring Fish Swarm Searching Algorithm for sequence-less CII able to generate optimal and feasible results for different test configurations.

In this paper, a renowned metaheuristic algorithm named chemical reaction optimization (CRO) is applied to solve the resource constrained project scheduling problem (RCPSP). This work employed chemical reaction optimization to schedule project tasks to minimize makespan concerning resource and precedence constraints. Chemical reaction optimization is a population-based metaheuristic algorithm. CRO is applied to RCPSP by redesigning its basic operators and taking solutions from the search space using priority-based selection to achieve a better result. The proposed algorithm based on CRO is then tested on large benchmark instances and compared with other metaheuristic algorithms. The experimental results have shown that our proposed method provides better results than other states of art algorithms in terms of both the qualities of result and execution time.

Path planning is an important aspect of an autonomous Unmanned Ariel Vehicle (UAV). As finding the best path is a non-deterministic problem, meta-heuristic algorithms proved to be a better choice in recent years. Particle Swarm Optimization (PSO) is one of the widely applied meta-heuristic algorithms for non-deterministic problem due to simplicity and ease of implementation. However, the lack of diversity in the particles in PSO algorithm generates a low-quality path for UAV. In this paper, we presented a modified PSO algorithm called n-PSO. In the algorithm, a dynamic neighborhood approach is proposed to improve the diversity of the particles. The n-PSO algorithm is applied to UAV path planning and simulated in a 3D environment. We compared the algorithm with two widely used versions of PSO for UAV path planning. The proposed algorithm showed significant improvement in particles diversity that plays an important role to produce better UAV path. At the end, we presented a time cost analysis of the algorithm for UAV path planning.

Solving quadratic equation efficiently is a real-world challenge nowadays, due to its wide applications in the task of determining a product’s profit, calculating areas or formulating the speed of an object. The general approach of finding the roots of a quadratic equation is not enough efficient due to the requirement of high computation time. Because of the Genetic Algorithm’s stochastic characteristics and efficiency in solving problems it can be used to find roots of quadratic equation precisely. In modern athletics reducing the computation time of solving the quadratic equation has been so inevitable where using a genetic algorithm can find a quick solution that doesn’t violate any of the constraints and with high precision also. Optimization has been done in the Crossover and Mutation process which has reduced the number of iterations for solving the equation. It reduces the time complexity of the existing approach of solving the quadratic equation and reaches towards the goal efficiently.

Link prediction is widely used in network analysis to identify future links between nodes. Link prediction has an important place in terms of being applicable to many real-world networks with dynamic structure. Networks with dynamic structure, such as social networks, scientific collaboration networks and metabolic networks, are networks in which link prediction studies are performed. In addition, it is seen that there are few studies showing the feasibility of link prediction by creating networks from different areas. In this study, in order to show the applicability of link prediction processes in different fields link prediction was made by applying traditional link prediction methods in the networks formed from the data of football competitions played after the groups between the years 2004–2017 in the UEFA European League. The AUC metric was used to measure the success of forecasting. The results show that link prediction methods can be used in sports networks.

Rice production assumes the most noteworthy part in national economy of Bangladesh. But due to several weather circumstances, rice production is being influenced day by day. In this research work, present a sustainable rice production analysis and forecasting rice yield for Aus, Aman and Boro rice based on weather circumstances. This paper aims to forecast rice production on the basis of weather parameters (Temperature, Rainfall, Humidity) and then predict future rice production based on previous data analysis. This research work has considered here Multiple Linear Regression, Support Vector Machine and Random Forest methods of data mining for selected region of Bangladesh. On the basis of the final calculating result, the analysis will help the farmers to understand which types of rice will be planted in which weather and it will help to achieve greater profit in the economy of Bangladesh.

Stock price forecasting, is one of the most significant financial complexities, since data are not reliable and noisy, impacting many factors. This article offers a machine learning model for the stock price prediction using Support Vector Machine-Regression (SVR) with two different kernels which are Radial Basis Function (RBF) and linear kernel. This study shows the Prediction and accuracy comparison between Support Vector Regression (SVR) and Linear Regression (LR) and also the accuracy comparison for different kernels of Support vector Regression (SVR). The model has used sum squared error (SSE) to determine the accuracy of each algorithm; which has shown significant improvement than the other studies. This analysis is conducted on the price data of about five years of Grameenphone listed on Dhaka Stock Exchange (DSE). The highest accuracy was found with Linear Regression model in every case with the highest accuracy of about 97.07% followed by SVR (Linear) model and SVR (radial basis function) model with the highest accuracy rate of about 97.06% and 96.82%. In some cases the accuracy of SVR (radial basis function) was higher than SVR (linear). But it was the Linear Regression which had the highest accuracy of all in every case.

Cross-project defect prediction (CPDP) is using due to the limitation of within project defect prediction (WPDP) in Software Defect Prediction (SDP) research. CPDP aims to train one project data to predict another project using the machine learning technique. The source and target projects are different in the CPDP setting, because of various structured source-target projects, sometimes it may not be a perfect combination. This study represents a categorical data set ensemble technique, where multiple data sets have been aggregated for source data instead of using a single data set. The method has been evaluated on nine data sets, taken from the publicly accessible repository with two performance indicators. The results of this data set ensemble approach show the improvement of the prediction performance over 65% combinations compared with traditional CPDP models. The results also show that same categories (homogeneous) train-test data set pairs give high performance; otherwise, the prediction performances of different category data sets are mostly collapsed. Therefore, the proposed scheme is recommended as an alternative to predict defects that can improve the prediction of most of the cases compared with traditional cross-project SDP models.

Software Defect Prediction (SDP) is a popular research area which plays an important role for software quality. It works as an indicator of whether a software module is defect-free or defective. In this study, a review has been conducted from January 2015 to August 2019 and 165 articles are selected in the area of SDP to know the prevalence of Machine Learning (ML) techniques. These articles are collected by searching in Google Scholar, and they are published in various platforms (e.g., IEEE, Springer, Elsevier). Firstly the information has been extracted from the collected particles, and then the information has been pre-processed, categorized, visualized, and finally, the results have been reported. The result shows the most frequently used data sets, classifiers, performance metrics, and techniques in SDP. This investigation will help to find the prevalence of ML techniques in SDP and give a quick view to understand the trends of ML techniques in defect prediction research.

Nowadays, Software Process Improvement popularly known as SPI has been able to receive an immense concern in the continuous process to purify software quality. Several Agile methodologies previously have worked with Extreme programming (XP). Before improving the process, defect prevention (DP) is inevitable. In addition, DP largely depends on defect detection either found earlier in the design and implementation stages or held in the testing phases. However, testing maturity model integration (TMMI) has a crucial aspect in DP as well as process improvement of the software. In particular, when software gets validated by being tested and fixed the defects up, it achieves the maximum capability maturity model integration (CMMI) aiming the process improvement. Here, the article has proposed an improved defect detection and prevention model to enhance the software process following the approach of XP. Besides, as a unique contribution, we have united the capability and testing model integration to ensure better SPI.

FIFA world cup is the most prestigious football tournament and widely viewed sporting event in the world. People support different teams (countries) of FIFA world cup based on players’ skills, number of winning trophies, and deliberate strategies that are applied by these teams during the tournament. These people share their opinion, criticism, love, and affection on the social media, i.e., Twitter. In this paper, we predict users’ FIFA world cup supporting preference from their tweets. First, we analyze user’s tweets and build two different types of classifiers by using LIWC and ELMo Word Embedding based techniques. These classifiers predict which team a user prefers from her word usage pattern in tweets. We find that Random Forest classifier performs the best for LIWC based model. We also find deep learning based word embedding technique, ELMo, achieves decent potential to predict users’ team supporting preference. Later, we build a multi-level weighted ensemble model to integrate both of the independent models, i.e., LIWC and ELMo. Our ensemble model shows substantial prediction potential (average accuracy-83.5%) to predict users’ FIFA world cup supporting preference from their tweets.

Machine learning method gives a learning technique that can be applied to extract information from data. Lots of researches are being conducted that involves machine learning techniques for medical diagnosis, prediction and treatment. The goal of this study is to perform several machine learning actions for finding the appropriate mode of birth (cesarean or normal) to minimize maternal mortality rate. To generate a computer-aided decision for selecting between the most common way of baby birth, C-section and vaginal birth, we have used supervised machine learning to train our classification model. A dataset consists of the information of 13,527 delivery patients has been collected from Tarail Upazilla Health complex, Bangladesh. We have implemented nine machine learning classifier algorithms over the whole datasets and compared the performances of all those proposed techniques. The computer recommended mode of baby delivery suggested by the most convincing method named “impact learning,” showed an accuracy of 0.89089172 with the F1 value of 0.877871741.

This paper aimed to classify two forearm muscles known as Flexor Carpi Ulnaris (FCU) and Extensor Carpi Radialis Longus (ECRL) using surface Electromyography (sEMG) signal during different hand prehension tasks, such as cylindrical, tip, spherical, palmar, lateral and hook while grasping any object. Thirteen Machine Learning (ML) algorithms were analyzed to compare their performance using a single EMG time domain feature called integrated EMG (IEMG). The tree-based methods have the top performance to classify the forearm muscles than other ML methods among all those 13 ML algorithms. Results showed that 4 out of 5 tree-based classifiers achieved more than 75% accuracies, where the random forest method showed maximum classification accuracy (85.07%). Additionally, these tree-based ML methods computed the variable importance in classification margin. The results showed that the lateral grasping was the most important moving variable for all those algorithms except AdaBoost where tipping was the most significant movement variable for this method. We hope, this ML- and EMG-based classification results presented in the paper may alleviate some of the problems in implementing advanced forearm prosthetics, rehabilitation devices and assistive biomedical robots.

Bangladesh is a frequently affected by river flood and flash flood because of its geographical location. Along with the number of vulnerabilities, flood is cause sever health related problems. Thus objective of this study was to develop a prediction model for self-assessed health for the people of flood-prone area of Bangladesh. A CHAID technique is applied to predict the self-assessed health status. Data was collected from 883 individuals who were selected applying multistage random from four selected flood affected districts - Sunamgonj, Chattogram, Jamalpur and Gaiandha of Bangladesh. It is observed that more than 54% people of the flood affected area had reported that they were in poor health condition. In addition, food scarcity, worried about future, health awareness, use of hygienic toilet and education level were found the influential factors for self-assessed health status. However, food scarcity was the most influential factors for the prediction model. Accuracy, Precision, Recall and F1 Score for the training model were found 75.1%, 82.01%, 74.5% and 78.1% respectively whereas for test model were 74.1%, 85.5%, 71.0% and 77.6% respectively. The prediction model would assist to identify people who might be under risk in the flood affected area and also can mitigate health related disaster in the area.

The aim of this review study is to analyze the techniques for predicting the surface EMG activities on upper limb muscles using different machine learning algorithms. In this study, we followed a systematic searching procedure to select articles from four different online databases, i.e. PubMed, Science Direct, IEEE Xplore and Biomed Central (published years between 2010 and 2018). In our searching procedure, we searched by characteristically with two keywords (“EMG” and “Machine Learning”) in the above four listed databases to find the related articles in the field of machine learning techniques for predicting surface EMG activities on upper limb muscles. From the searching of this review, we selected total 25 articles for predicting surface EMG signals on upper limb muscles, where 10 articles are provided most efficient and effective classifier of surface EMG signals, 11 articles described different hand gesture recognition using machine learning algorithms, 2 articles explained that the importance of muscles selection, 1 article presented the natural pinching technique and 1 article focus on evaluation error rate of movements. This review presents not only the machine learning techniques for prediction of surface EMG activities on upper limb muscles but also it focuses on the challenge of the machine learning techniques for predicting surface EMG data. In addition, we believe that this review also provides muscle related issues that will impact the prediction of surface EMG activities on muscle.

Cancer incidences are increasing day by day and has become a global burden now. The incidences are frequently occurring in low-income countries like Bangladesh. A retrospective descriptive type of study had been carried out by us over 500 patients of hematological cancer in Bangladesh. The “French American British” classification of hematological cancer is used to carry out the morphological typing. We have correlated Hematological Malignancy (HM) with occupation, residence, associated infection, idea about cause of cancer and previous cancer history in relative. The analysis showed the diagnosed cancers are positively co-related with age, gender, occupation & idea about cause of cancer and negatively co-related with division, previously cancer history in relatives & associated disease. Our study shows the risk factor and the distribution pattern of hematological malignancy in the area of Bangladesh. It presents the distribution pattern of HM according to Age, Gender & correlation of HM with occupation, residence & other factor.

Breast cancer is the most common cancer among women and a major cause of death globally. A high percentage of the cancer death can be reduced if it is detected earlier (Stage 1 or 2). Early and non-invasive (health-friendly) diagnosis is the most essential key to detect breast cancer that ensures a fast and effective treatment for reducing women mortality. Ultra-wide band (UWB) technology is considered as an effective technique for breast cancer detection due to its health friendly (non-ionizing) nature to human tissue. The UWB technology uses the scattering or reflected wave/signal from breast tissue for diagnosis. A high-performance bio-antenna plays an important role in transmitting and receiving the UWB signal for this case. In this paper, breast cancer detection performance comparison of two types of UWB bio-antennas (pyramidal shaped UWB patch and the proposed modified T shaped UWB patch) has been investigated depending on accuracy. A system has been developed using a pair of UWB transceivers with bio-antennas and artificial neural network (ANN). The signals are transmitted and received through breast phantom for different arbitrary tumor size and location for considered antennas. The obtained tumor/cancer location and size detection accuracy are approximately 90.27% and 89.91% for pyramidal shaped antenna, whereas, those for the proposed (modified T shaped) antenna are nearly 91.03% and 91.09% respectively. The proposed (modified T shaped) antenna is comparatively better to detect early breast cancer than pyramidal shaped antenna, by showing its suitability for practical use in near future.

This paper represents a novel strategy for developing a disease diagnosis gadget from a patient’s statement. For that, the system solely accepts patients’ statements in a natural language like English and analyzes the patients’ statements to prognosis the symptoms the affected person is presently suffering from. The framework forms the patients’ discourse and afterward utilizes Term Frequency (TF) to find the indications of a malady. Cosine Similarity is utilized to settle on a final decision with respect to regarding disease diagnosis task. Cosine Similarity quantifies the similitude between two non-zero vectors in a vector space model where one of the vectors is constructed with the symptoms the patient is encountering and the rest is developed during knowledge base setup. The framework is tested over 1013 patients with various ailments and its accuracy up to 98.3%.

Diabetes is a silent-killer disease throughout the world. It is not curable, therefore, regular blood glucose concentration levels (BGCL) monitoring is necessary to be healthy in a long run. The traditional way of BGCL measurement is invasive by pricking and collecting blood sample from human arm (or finger-tip), then measuring the level either using a glucometer or sending to laboratory. This blood collecting process produces significant discomfort to the patients, especially to the children with type-A diabetes, resulting increased undetected-cases and health-complications. To overcome this drawbacks, a non-invasive ultra-wideband (UWB) BGCL measurement system is proposed here with enhanced software module. The hardware can be controlled through the graphical user interface (GUI) of software and can execute signal processing, feature extraction, and feature classification using artificial intelligence (AI). As AI, cascade forward neural network (CFNN) and naïve bayes (NB) algorithms are investigated, then CFNN with four independent features (skewness, kurtosis, variance, mean-absolute-deviation) are found to be best-suited for BGCL estimation. A transmit (Tx) antenna was placed at one side of left-earlobe to Tx UWB signals, and a receive (Rx) antenna at opposite side to Rx transmitted signals with BGCL marker. These signals are saved and used for AI training, validation and testing. The system with CFNN shows approximately 86.62% accuracy for BGCL measurement, which is 5.62% improved compared to other methods by showing its superiority. This enhanced system is affordable, effective and easy-to-use for all users (home and hospital), to reduce undetected diabetes cases and related mortality rate in near future.

In the area of sentiment analysis and classification, the performance of the classification tasks can be varied based on the usage of text vectorization and feature extraction methods. This paper represents a detailed investigation and analysis of the impact on feature extraction methods to attain the highest classification accuracy of the sentiment from user reviews. Unigram, Bigram and Trigram are applied as n-gram vectorization models with TF-IDF features extraction method individually. Accuracy, misclassification rate, Receiver Operating Characteristics (ROC) and recall-precision are used in this study to evaluate which are counted as the most important performance measurement parameters in machine learning based approaches. Parameters are measured by the output obtained from Bagged Decision Tree (BDT), Random Forest (RF), Ada Boost (ADA), Gradient Boost (GB) and Extra Tree (ET). The outcomes of this study is to find out the best fitted combination of term frequency–inverse document frequency (TF-IDF) and n-grams for different data size.

Recent years have seen rapid growth of research on sentiment analysis. In aspect-based sentiment analysis, the idea is to take sentiment analysis a step further and find out what exactly someone is talking about, and then measuring the sentiment if she or he likes or dislikes it. Sentiment analysis in Bengali language is progressing and is considered as an important research interest. Due to scarcity of resources like proper annotated dataset, corpora, lexicon such as part of speech tagger etc. aspect-based sentiment analysis hardly has been done in Bengali language. In this paper, we have conducted our experiments based on a recent work from 2018 using conventional supervised machine learning algorithms (RF, SVM, KNN) to perform one of the ABSA’s tasks - aspect category extraction. The work is done on two datasets named – Cricket and Restaurant. We then compared our results with the existing work. We used two traditional steps to clean data and found that less preprocessing leads to better F1 Score. For Cricket dataset, SVM and KNN performed better, resulting F1 score of 37% and 27%. For Restaurant dataset, RF and SVM achieved improved score of 35% and 39% respectively. Additionally, we selected two more algorithms LR and NB, LR achieved best F1 score (43%) for Restaurant dataset among all.

An accurate speech recognition system requires close observation of the selection of an error-free speech feature extraction model. This paper describes a prominent solution to obtain robust features from the sound spectrum and ensures the easy recognition of speech. The proposed architecture uses Tangent based (TB) auditory feature extraction that aims to find and process robust features from the sine wave of auditory signal data. This experiment suggests that every specific tune carries distinguishing signal patterns in the spectrum diagram and hence does the tangent of the amplitude of the same signal. To recognize the sound, a single attribute had been used rather than using multiple attributes where the slope of the sound spectrum being calculated.

Finding the semantic similarity between texts is not trivial and is an indispensable task in many NLP and information retrieval tasks. In this paper, we introduced a semantic sentence modeling approach for learning the similarity between sentences using long-short-term-memory (LSTM) networks. First, sentences are represented with high dimensional vectors based on the word-embedding model that encodes the semantic meaning of the sentences. Then the encoded sentences are used to train the siamese LSTM model. The trained model builds a structured high dimensional space and can predict the semantic similarity between sentences. We applied our proposed method on two benchmark datasets on semantic textual similarity. The experimental results exhibited the efficiency of our method and outperformed some know related methods with 82% and 83% accuracy in terms of Pearson’s r and Spearman’s $$\rho $$ , respectively.

Natural language processing is the current topic due to many important tasks like document classification, named entity recognition, opinion mining, sentiment analysis, textual entailment, etc. Such types of task in the Bangla language is also important. This research work endeavored to find out the word embedding of the Bengali language. Leveraging the fastText word embedding, it has shown significant performance in Bangla document classification without any prepossessing like lemmatization, stemming, and others. For the extrinsic evaluation of our word vectors, a classification problem-solving strategy has been used which showed an outstanding result. In the classification module, attempts have been made to classify 40 thousand News samples into 12 categories. For this purpose, three deep learning techniques have been used: Convolutional Neural Network (CNN), Bi-Directional LSTM (BLSTM) and Convolutional Bi-Directional LSTM (CBLSTM) alongside fastText. From the analogous study of all the parameters of every classifier implemented here, we found that the BLSTM technique is the most promising technique for this task. This technique achieved 91.49%, 87.87%, and 85.5% accuracies for Training, Testing, and Validation set, respectively.

Question Classification (QC) system classifies the questions in particular classes so that Question Answering (QA) System can provide correct answers for the questions. We present a two stage QC system for Bengali. One dimensional convolutional neural network (CNN) based model has been constructed for classifying questions into coarse classes in the first stage which uses word2vec feature representation of each word. A smart data balancing technique has been implemented in this stage which is a plus for any training dependent classification model. For each coarse class classified in the first stage, a separate Stochastic Gradient Descent (SGD) based classifier has been used in order to differentiate among the finer classes within that coarse class in stage two. TF-IDF representation of each word has been used as feature for each SGD classifier separately. Experiments show the effectiveness of this two stage classification method for Bengali question classification.

Authorship attribution is the process of identifying the probable author of an unknown document. This paper proposes a neural network based framework, which identifies the authorship from Bengali literary documents. For this purpose, a corpus consisting of 12,142 text documents of 23 writers/bloggers is built. A static dictionary is used to count vectorization and important features are selected using information gain. The proposed system is trained with 9099 documents and tested with 3043 documents. The experimental result shows that neural network with n-gram and parts of speech (PoS) features achieved 94% accuracy on developed corpus.

Multi-label text classification has been a key point of research in the area of text classification latterly. But to the best of our knowledge, there have been very few research on multi-label text classification for Bangla text. There is also inadequacy of proper dataset for multi-label classification on Bangla text. Multi-label classification has many applications in the real world. One of them is automated labeling of articles of online news portals so that readers can easily look up other news articles on similar topics by clicking on hyperlinks. We applied supervised multi-label classification techniques on Bangla news articles for automated tag generation to predict related topics. We have built a new dataset from scratch and applied various problem transformation methods for multi-label classification with naive bayes classifier, logistic regression and SVM. We have analyzed the performance of these algorithms on Bangla news articles with precision, recall, f1-score and hamming loss. The dataset and the analysis of the results can be valuable for further research on multi-label text classification of Bangla text. We have open-sourced the dataset and the source code of this work ( http://bit.ly/34cSNCR ).

WCE (Wireless Capsule Endoscopy) has become one of the most significant inventions for detecting different types of digestive tract diseases of humans. Distinct types of abnormalities like polyps, ulcer, tumor and intestine cancer are diagnosed by the clinicians with the implement of WCE in a convenient way. In order to deduce the incubus of the physicians an automated and efficient recognition system is required. In this paper, an advanced method for automatically detecting ulcers in the images of the WCE video record is proposed using the HSV color model. Region of interest (ROI) was identified applying a threshold on images that were extracted from the video of WCE. Local features have been extracted only from the ROI which is usually a small part of an image that offers a low computational cost. Linear discriminant analysis has been used for the separation of ulcer and non-ulcer images. The proposed algorithm was tested on a publicly available database. The performance has obtained accuracy 87.55%, sensitivity 94.70%, specificity 83.30%, precision 75.00% and F1 score 83.70%. Hence, the proposed method outperforms an efficient method that will create a great impact in this research arena.

At present age estimation and gender classification task has achieved a great importance due to analyzing the category of people in social media, business, customers’ choice etc. Automatic age and gender classification task from analyzing facial images has become a concern to this competitive world. In this paper, we have proposed to apply transfer learning technique on facial images of people of different ages and gender. Age and gender are special attributes which can be extracted from facial images. A deep convolutional neural network is trained using our target dataset to achieve a good classification performance. We have evaluated the classification performance on Adience benchmark for age and gender estimation using ResNet50, VGG16, VGG14 and VGG17 deep CNN models. Using an ensemble technique (majority voting) of these (VGG) classifiers, we have found approximately 90% classification accuracy on age estimation task. We have also found 94% 1-off age classification accuracy using VGG14.

At present, cancer is a second leading cause of death which rises the global burden. Among them acute lymphoblastic leukemia is a subtype of blood cancer which is most common in child as well as adults. It occurs when the number of lymphoblast is more producing from stem cells. Over time the accumulation of this abnormal cells in bone marrow prevents to produce other healthy blood cells in our body which is very dangerous. So, early detection is one of the most important which can increase patient’s survivability and treatment options. For cancer diagnosis, Ultrasound, Mammogram, MRI and microscopic images are some common methods used in medical science. Some basic detection processes of leukemia are CBC, PBS test and bone marrow test based on microscopic images. For blood cancer diagnosis, microscopic images are used manually which is time consuming and less accurate and can produce non standardized reports. So, it needs to detect leukemia automatically. Recently some computer aided methods are generated to diagnosis leukemia which are more reliable, more accurate, more precise and faster than manual diagnosis methods. In this paper a new automatic system has been proposed to detect all based on several image processing techniques from microscopic image of blood smear such as, segmentation, preprocessing, enhancement for getting better performance. To, classify blast cells and healthy cells ensemble classifier has been used with several types of feature such as, texture features, geometric features, statistic features. In this paper 99.1% accuracy, 98% Sensitivity have been achieved.

Software Defined Networking (SDN) approach is a generalized concept that de-couples software plane from the hardware plane of a network. SDN can be an alternative to well-defined protocol stack, scalability, and full resource management capabilities. SDN for wireless environment became a popular research field for future deployment. Therefore, performance issues of Software Defined Wireless Networking (SDWN) have become important in order to study and analyze the underlying network design, scope, and capabilities. This research work represents the performance analysis of SDWN for multiple domains and inter-controller communication. Integration of IP mobility with the Mobile Nodes (MN) affects TCP throughput, bandwidth, transmission jitter and latency of underlying SDWN. This paper concludes that SDWN has both integrated performance in efficient handoff through IP mobility solution and somewhere penalties as well in terms of inter-controller communication. In the end, a comparative study with distributed mobility solutions is performed against an IP based model.

The Internet of Things (IoT) is an arrangement of connected numerous digital devices usually contained Unique Identifiers (UIDs) and have the capability to exchange data over a network without any human interaction. Another new paradigm Software-Defined Networking (SDN) comes in for the organization and control of the large amount of data produced by IoT devices. It separates the data plane from the control plane of network devices which enables easy configuration and management of those devices. Furthermore, Network Function Virtualization (NFV) is emerged to optimize and secure the SDN-IoT network. It enables network devices to be deployed as virtualized components via software. In this research, the authors have proposed an SDN based distributed IoT network with NFV implementation for smart cities. Where smart city is a residential area which utilizes Information and Communication Technology (ICT) as well as IoT network to develop the standard of living of its residents. The integration of NFV in the SDN-IoT network improves the network performance by increasing throughput, and time sequence while mitigating the round trip time as well. Moreover, the authors have used multiple distributed controllers and a clustering scheme to improve load balancing, scalability, availability, integrity, and security of the whole network.

Delay-Tolerant Network (DTN) is a resource-bound networking system which consists of many intermittently connected, movable devices known as nodes. Energy can be considered as an important resource for DTN scenarios since these nodes have limited energy. In order to perfect network enforcement, it is necessary to exploit the energy of the nodes efficiently. In DTN, most of the node energy is consumed because of mobility, scanning neighbors to deliver message and message transmission. Node energy has a significant role for successful transmission of messages. Higher energy of a node means that it has a high possibility to route its message with success across the network. So, for effective message routing it is mandatory to select an energy efficient routing mechanism in DTN environment. This point makes us interested to study the consumption of node energy in DTN scenarios. Within this research, the study of energy issue is focused for DTN routing approaches: Epidemic, Resource Allocation Protocol for Intentional DTN (RAPID), MaxProp, Probabilistic Routing Protocol using History of Encounters and Transitivity (PRoPHET), Spray and Wait, and Spray and Focus with their comparative performance analysis on behalf of four performance criteria: average remaining energy of node, delivery ratio, average latency, and transmission cost. Simulations are performed in Opportunistic Network Environment (ONE) simulator by varying node density while keeping message Time-To-Live (TTL) fixed and further, message TTL is changed while node density is kept fixed. We have found that Spray and Wait is the most energy efficient DTN routing scheme, whereas Spray and Focus yields the best performance in terms of delivery ratio, average latency and transmission cost.

This research introduce a robust optimization model to reduce the congestion ratio in communications network considering uncertainty in the traffic demands. The propose formulation is depended on a model called the pipe model. Network traffic demand is fixed in the pipe model and most of the previous researches consider traffic fluctuation locally. Our proposed model can deal with fluctuation in the traffic demands and considers this fluctuation all over the network. We formulate the robust optimization model in the form of second-order cone programming (SOCP) problem which is tractable by optimization software. The numerical experiments determine the efficiency of our model in terms of reducing the congestion ratio compared to the others model.

This paper proposes a mobile application named “Sightless Helper”, for assisting blind or visually impaired people. The application uses footstep counting and GPS for indoor and outdoor navigation. It can detect objects and unsafe areas to ensure safe navigation. The system consists of voice recognition, touchpad, button and shaking sensor for easy interaction between the user and the system. During any kind of accident, it can detect unusual shaking of the user, and send his/her location to some emergency contacts. “Sightless Helper” pro-vides several useful additional features such as calendar, news reading, barcode reading, battery monitoring, etc. The performance of the application is tested considering voice recognition time and location sending time. The experimental result shows that the voice recognition time of the application is around 6.303 ms and 6.375 ms for male and female voices respectively. The average location sending time is nearly 7.629 ms to any distance. The usability test result reveals that the proposed application has an average 72.2% System Usability Scale (SUS) score, showing its suitability for practical implementation.

In improvement of the quality of health care services, Internet of Things (IoT) has evolved rapidly for monitoring patient from distance. However, notifying health status based on continuous change of health condition for immediate healing to patient, existing systems has some limitations. In this paper, we demonstrate a smart health monitoring technology for diabetic patients which follows up their health condition depending on sugar level, heart pulse, food intake, sleep time and exercise. To illustrate, this technology takes the variables (data) as input through sensors continuously and process with neural network to evaluate the data, resulting four modes of health risk status: low, medium, high and extreme. The range of the risk status can differ based on patient’s type and previous histories of their health. In addition, an automatic phone call and/or SMS notification is being sent to patient’s relative along with patient’s location if his/her health condition is at high or extreme risk. Besides, it also calls patients nearest hospital in case of extreme risk. However, the system provides allied instruction as voice command to patient’s mobile in both cases. This technology has been experimented on 25 diabetic patients successfully and achieved 84.29% accuracy to identify the proper risk level, which is a highly acceptable level of identifying health risk status.

Artificial Intelligence (AI) based recommender systems help to make our life easy and comfortable. From simple chatbot to YouTube recommendation, AI is used to recommend news, videos, etc. which provide us more information and saves our time. In big cities, parking seems to be a major problem where commuters need to find a suitable parking space among many parking areas which cause wastage of time and fuel. Our paper proposes a parking recommender system where commuters will be suggested a parking area to a nearby place for helping them to save time, parking cost and ensure high security. To collect data of parking spaces, we propose a Cloud architecture where we use the concept of Edge and Cloud computing to collect and process data smoothly and reduce latency. To deal with bigger amounts of data we use Data Streaming Pipelining to process and analyze those data. We use Amazon Web Services (AWS) to implement our proposed Cloud architecture. For creating the AI based recommender system, we propose the Q-learning algorithm with $${\varepsilon }$$ -soft policy to suggest nearby parking areas. Our novel approach will be helpful for both local and global citizens to find an ideal parking area close to their working place, home, etc. Our proposed Cloud architecture is able to reduce latency and make data transferring system faster. Also the Q-learning algorithm can outperform in terms of both certain and uncertain situations.

At present, neural style transfer technique is gaining more and more popularity in different fields like sports and entertainment. During the artistic style transfer of an image, style loss and content loss occurs. Several research works are being performed in present with a view to reduce the rate of both style and content loss, but some of these processes require much time for style transfer. In this paper, we propose a style transfer method using convolutional neural network that minimizes the rate of style and content loss in a minimal time compared to some other works. The proposed method is applied on different images and artworks. The results are compared with some recent research works related to this method and the proposed method is found to be 5–7% more efficient and faster than those related works.

Blockchain has emerged as a revolutionary technology that has been envisioned to disrupt several industries, including financial domains with its decentralised and highly-secure design. However, since the beginning of its evolution, it has been highly criticised for the difficulties in dealing with it properly. As cryptocurrency is the most successful application of blockchain, we aim to identify the potential obstacles and usability issues, which might hinder its wide-scale adoption, of five applications (i.e., wallets) that are used to manage cryptocurrencies. Applying the analytical cognitive walk-through usability inspection method, we investigate common usability issues with desktop and mobile-based wallets. Our results reveal that both wallets lack good usability in performing the fundamental tasks which can be improved significantly. We summarise our findings and point out the aspects where the issues exists so that improving those areas can result in better user experience and adoption.

Data cubes come up with a suitable paradigm for storing, accessing, processing and analysis multidimensional data. Conventional Multidimensional Arrays (CMA) are the basic data structure to process such multidimensional data. But the performance of the MDAs degrades when the number of dimension increases. In this paper, we propose a new approach for computing multidimensional data cube using conversion of dimensions of the multidimensional array. We design efficient algorithms for Multidimensional On Line Analytical Processing (MOLAP) operations using the Converted two dimensional Array (C2A). We represent the MOLAP array as a Converted two dimensional Array where n-dimension is converted into two dimension. Then we apply the operations of data cube namely slice and dice on both CMA and C2A. We calculate the time for slice and dice operations for CMA and C2A. The proposed model requires less time for index computation when number of dimension is high. The cache miss rate is also lower for C2A based implementation. Therefore, our proposed algorithm shows superior performance than the traditional scheme.

In this letter, a bow-tie-type photonic crystal fiber (PCF) with high birefringence (Hi-Bi) has been proposed. The core of the PCF is elliptical with Chalcogenide glass ( $$Ga_{8}Sb_{32}S_{60}$$ ) material. The whole analysis of the PCF is finished by the finite element method (FEM) for wavelength ranging from 2,000 nm to 3,000 nm to obtain some optical parameters like birefringence, beat length, power fraction, numerical aperture, effective refractive area, and nonlinearity. Therefore, a perfectly matched layer (PML) is also used to throw out unwanted radiation directed as an absorbing boundary condition (ABC). It has generated high birefringence (Hi-Bi) of 0.287 at 2,975 nm wavelength, the highest power fraction of 89.39% at 2,000 nm wavelength, the higher numerical aperture of 0.86, and the better nonlinearity of 6.10 $$\times $$ $$10^{3}$$ $$\mathrm{W}^{-1} \mathrm{Km}^{-1}$$ . Hence, the proposed PCF plays a significant role in PCF areas with the better polarization filter, cross talk (CT), sensing, and nonlinear applications.

Cancer is a disease involving the uncontrollable growth of cells with potential strafe to other organs of the body. Stress is a state of the body a non-specific response to any demand for change. Cancer had a deep relation with stress. Activation of the stress response and exposure to the associated hormones could promote the growth and spread of tumors. The immune system can be important for finding and eliminating cancer cells. This study is based on Cancer and Stress. In this study, we collect responsible genes from NCBI’s Gene database individually for stress and cancer. After that, common responsible genes were collected by using Venny online tools. From the common genes, we had constructed a protein-protein interaction network using the STRING database. Afterward, the top 10 hub genes were identified by using CytoHubba. Hub genes were identified based on their degree value where degree value more than or equal 72 are considered as hub gene. These hub genes may use to design a potential drug for cancer and stress combine. We have collected 3264 and 9433 human genes for Cancer and Stress respectively. 2477 common genes are found through Venny. We have been identified the UBC, TP53, RPS3, RPL5, RPL11, RPS27A, RPL19, RPL3, RPS7, and CTNNB1 as targeted hub genes by using the CytoHubba plugin of Cytoscape.

Data have ascended a place among capital, labor and land in production. The emerging data-driven economy has facilitated the scope of growth of data-entry industry – an industry equipped with modern computing and communication infrastructure enriched with specialized software interface allowing data-entry professionals to look into the source, collect and store target data. These data are used in business intelligence and analytics for value creation. By nature, data-entry is a tedious and repetitive task. It not only hampers creativity of the operators but also leave a possibility of wrong entry. In this paper, an innovative algorithm has been proposed which can automate the date entry industry with above 97% accuracy, more than 15 times faster than existing speed with no additional cost apart from the cost of existing infrastructure. The proposed algorithm has been tested and compared with several data-entry focused companies which demonstrate that it outperforms current manual data-entry approach and it has the potential to revolutionize the data-entry industry.

Machine learning methods such as deep neural networks have remarkably improved plant species classification in recent years. It is very challenging task to classify plant species based on their categories. In this work, deep learning approach is explained to identify and classify succulent plant species using VGG19, three layers CNN and five layers CNN network on our dataset. The proposed architecture achieved a significant result from VGG19 and three layers CNN model. In succulent plant image dataset, there are 10 different classes of succulent and non-succulent plants. The dataset consists of 3632 succulent plant images and 200 non-succulent plant images. The model achieved 99.77% accuracy which performs better than VGG19 and three layers CNN model.

From the last few decades, smoke detection performed for noble purposes like to rescue people from fire, make wood-land or wildlife safe from fire disaster and so on. Most of those detections were sensor based where detectors detect smoke optically or by physical processes and which causes false alarm most of the time. By the passing time, the author’s tries to overcome those false alarm rates by introducing hand-featured methods. From this perspective, those established systems performed better than sensor based tools. However, coming towards a significant point, in most instances, only one or two certain areas like forest were considered in addressing smoke. Now, moving on this research, we aimed to experience indeed with detecting diverse circumstances smoke by the Faster R-CNN approach based on the Inception-V2 deep neural network. We focused on the single class, i.e., smoke and training the method with images of our own combined extracted image frames. The proposed method achieves 97.31% detection accuracy and is compared to previous approaches to show higher detection accuracy over recent works.

With the increment of computation power, recognizing handwritten Character has become popular and significant improvement has been achieved for most of the major languages. But Bengali character recognition system is not well enough because of the presence of perplexing character and excessive cursive in its characters. Although several research works have been conducted for recognizing the Bengali characters, an efficient procedure is yet to discover. As the number of datasets is inadequate, most of these studies could not achieve a satisfactory level. So we propose here to train a Convolution Neural Network (CNN) and tune the parameters for better accuracy. This procedure is applied to CMATERDB 3.1.2 dataset with 15000.

Road surface monitoring is mostly done manually in cities which is an intensive process of time consuming and labor work. The intention of this paper is to research on road damage detection and classification from road surface images using object detection method. This paper applied multiple convolutional neural network (CNN) algorithm to classify road damage and discovered which algorithm performs better in road damage detection and classification. The damages are classified in three categories pothole, crack and revealing. For this research data was collected from street of Dhaka city using smartphone camera and prepossessed the data like image resize, white balance, contrast transformation, labeling. This study applies R-CNN and faster R-CNN for object detection of road damages and apply Support Vector Machine (SVM) for classification and gets a better result from previous studies. Then losses are calculated using different loss functions. The results demonstrate the highest 98.88% accuracy and the lowest loss is 0.01.