Bangladesh's economy, like many other contemporary nations, is fundamental to its overall structure. The stock market is one crucial and well-known component of this. Therefore, it is essential to anticipate it earlier to better prepare for market swings. The price of stocks is influenced by a wide range of factors, including geopolitical and economic ones. However, stock data and news sentiment are two that we can leverage ourselves in stock forecasting. Since news strongly influences the stock market, regular and institutional investors often base their decisions on news analysis. Integrating news sentiment analysis into a predictive model holds immense potential when building stock price forecasters. In our research, we have built a web application that reads data from the Dhaka Stock Exchange (DSE) and then relays the output prediction to the user. We use LSTM and GRU for the stock forecasting models on seven banks-four private and three public. For the sentiment analysis task for each of these seven datasets, we employ FinBERT. React, Flask, and MongoDB power our web application, which serves as the medium for all communication. Furthermore, we can confidently state that our research outcomes were precise and valuable regarding the future prices of Bangladeshi stocks. This paper will demonstrate and explain this endeavor's process and outcome trajectory.

Last-minute hotel reservation cancellations, often due to schedule changes, significantly impact hotel revenue. We can predict these cancellations by analyzing booking parameters such as the number of guests, rooms booked, and booking duration. Our study employed four models: Logistic Regression, SVM (Linear and RBF kernels), Decision Tree, and Random Forest. We evaluated these models using training and test datasets, with the Random Forest model achieving the highest accuracy at 99%. Additionally, we introduced a weighted average technique to help users visualize model performance and select the most suitable model for their needs.

In the modern age, with the help of social media, communication has become available for everyone. Offensive text is broadly used in social media to humiliate or threaten someone. Offensive text like a bully, trolling, threats, and sexual harassment are used to demotivate someone. We have gathered a dataset of 44,000 comments from social media. We use five different models: DistilBERT, Multilingual BERT, XLM-RoBERTa-base, XLM-RoBERTa-large, and BanglaBERT. Variations in various parameters, e.g., learning rate, dropout rate, training epoch, early stopping, and batch size, are made to get better results. From our proposed model, the XLM-RoBERTa-base shows the highest accuracy, 83.54%, whereas m-BERT provides the highest AUC value of 0.85.

It might be challenging to choose the best text preprocessing strategy in the field of natural language processing (NLP) due to the variety of techniques available. Given the popularity of transformer models, we wondered if preprocessing was necessary and, if so, what methods would improve the models' performance. Especially when working with clinical text data, accuracy is crucial. Our goal was to find an appropriate pre-processing pipeline for clinical texts that maintains or improves model performance. We experienced four common preprocessing techniques and their groupings on two datasets from MIMIC-3 and PubMed. We used four models: BERT base, BioBERT, BioClinicalBERT, and RoBERTa. The varied accuracy results from existing techniques inspired us to develop a new pipeline to improve accuracy. Our pipeline starts with removing repeated punctuation, normalizing the text with a CleanText function, and filtering less important words using TF-IDF scores to keep clinically applicable terms and moderate noise. Our results presented that our pipeline outperformed the base models. For the MIMIC-3 dataset, the BERT base model achieved 90.16% accuracy, and for the PubMed dataset, BioBERT achieved 64.20% accuracy. We also found that removing stop words decreased accuracy, while using TF-IDF either maintained or improved it up to 3%. Additionally, as we removed less important words from the documents our pipeline considerably reduced training time up to 17%.

Network and security is connected with each other. At present days thinking about communication without the network is impossible. Since the network is a public domain and anyone can use it, some corrupted people and hackers will try to gain profit by intruding others' sensitive information. The network intrusion can be done by some hackers or the network itself. For that reason ensuring security is more challenging than ever before. The proposed model detects the intrusion types currently in the network by using deep learning ANN and stack ensemble techniques. There is no space for compromise in security so it is strongly recommended to use an intrusion detection system that is more accurate and efficient. The reason for using a stacked ensemble is that even though a single deep learning model is strong enough to detect the intrusion, yet by using the stacked ensemble combines multiple deep learning models together to get a more efficient stronger intrusion detection mechanism.

IoT enabled devices are gaining huge momentum these days due to the rising importance for automation and interconnection among devices. This exponential growth has been mainly driven by the advancement in wireless communication, online storage, artificial intelligence and most importantly, high internet penetration among people. In this paper, a web-connected human heart rate monitoring system has been introduced. The system in question utilizes a concept known as photoplethysmography to register the heartbeats of an individual in a non-invasive way. Low-cost and commonly available electronic components are used to assemble the system. The sensing unit consists of a 5mm Infrared (IR) LED and a 5mm IR receiver. Data processing is carried out by an A VR microcontroller unit. Moreover, noise removal is achieved by the means of digital filtering method. The readings are then uploaded to an online server, which can then be viewed in a mobile phone application. The contribution of this research is two folds. First, to make health monitoring devices accessible to people in underdeveloped nations. Second, providing online connectivity will aid people living in remote areas where physical access to healthcare facilities or a doctor is difficult. Results show good agreement with actual heart rate reading. The analogue circuitry and the noise elimination technique described in this work can be extended to construct other similar health monitoring devices as well.

Depression has become a severe problem with grave consequences for Bangladesh's social and economic sectors. Fearing societal stigmas and misconceptions regarding depression, most individuals in our nation put off getting the medical care they need, which causes disaster in their lives. This is why it is difficult to gather accurate information on a depressed person, making it sometimes impossible to offer assistance. Social media can be a valuable source of information for recognizing depression among Bangladeshis due to its extensive use. We used a variety of natural language processing (NLP) techniques and machine learning algorithms (including LR, SVM, RF, GBDT, and XGBoost) to analyze postings from 1778 Facebook users (Female 25%, Male 60%, Other 15%) in this study to test the viability of determining the level of depression among locals of Chittagong and Dhaka over the previous four months. We found the RF depression classification model with the highest accuracy (0.7454) and F1-score (0.60) for detecting depression levels. To investigate the situation of depression at the local level, we also employed the medically approved Patient Health Questionnaire-9 (PHQ9) assessment on residents of Chittagong and Dhaka (both online and offline). By utilizing their comparison, we could determine the depression levels among individuals of various ages and sexual orientations in Dhaka and Chittagong.

Widespread floods are one of the most destructive natural phenomena that frequently occur in Bangladesh. It devastates human life, food security, shelter, and social and financial losses. Due to Bangladesh's current economic situation, installing and maintaining comprehensive flood gauges for national flood assessments will not be viable. Hence, online social media platforms can be a valuable avenue to get real-world data to perform flood assessments. This study investigates whether the abundance of data collected from social media in Bangladesh could be used to achieve a reliable and consistent flood assessment in Bangladesh. The data gathered from online platforms are related to flooding and exist in the form of videos, images, and text. The data collected is analyzed with a Machine Learning approach. The digital data of images and video frames were converted into numeric values using the VGG-16 architecture. A Convolutional Neural Network takes in the numeric data produced by the VGG-16 for classification. The classification made accuracy of 92% for the image-based model and 87% for the text-based model.

A farming method known as aquaponics aims to be a more effective response to the world's worries about food production and scarcity. In order to produce plants or vegetables in addition to fish to meet the rising global demand for food, this technique combines aquaculture (fish farming) and aquaponics (plants grown without soil). This technique or system promises to utilize less water, fertilizer, pesticides, and space while yet producing as much as conventional farming. The challenge is making it more usable and scalable for commercial and public use. To overcome such challenges, this technique must be intelligent and automated by rigorous sensing, monitoring, and system control. Our current research aims to develop a practical smart IoT-based aquaponics system. It offers a workable solution and ideas for further investigation while also looking for gaps in the earlier investigations.

E-commerce platforms have made our life easier and bought plenty of advantages too. However, due to fraud and scams, trust is a concern while buying products online. In this study, we proposed a blockchain-based architecture for the e-commerce sector where data mining technology is used to detect fraudulent users by generating precise and effective rules, and smart contracts are used for enforcement and functionality management within the blockchain network. Our data mining approach yielded a competitive accuracy, precision, recall, and f1-measure of over 99% compared to the state-of-the-art. High performing rules are further tested for ten simulated cycles under completely unseen data to test their rigidity in a real-time scenario. Details analysis of delineation (if any) in rule antecedents has also been analyzed for these cycles. Tested and selected rules are stored in smart contracts deployed within blockchain network for security and immutability. For ensuring strict order criteria maintenance, better return policy, authentic review and scam avoidance, two smart contracts have been written which implement seller reputation mechanism and authentic review maintenance while safeguarding the seller from intentional defaming too. Altogether, the goal of this study is to establish a balance between all the parties within an e-commerce platform so that everyone's right is protected, money is safe and resources are not exploited.

Large-scale management of surface water resources in urban areas can be difficult, especially if the region is subject to monsoonal waterlogging. Deep learning-based methods for computer vision tasks, such as image segmentation, can effectively be applied to remote sensing data for generating water body maps of large cities, aiding managerial entities, urban planners and policymakers. The robustness of these models to erroneous pixel level training class labels has not been studied in water body segmentation context. Label noise is commonly experienced in classification tasks and may hinder the performance. We collected and densely labeled Sentinel-2 images over Dhaka, one of the most densely populated and flood prone megacities in the world. We synthetically injected four types of label noise viz., i) Gaussian noise, ii) translation, ii) rotation, and iv) mirroring. Our primary objective is to observe and quantitatively analyze the effects of label noise on remote sensing data-driven deep learning models for water body segmentation. Our results show that salt and pepper noise (injected artificially using Gaussian noise) of only 50% can cause a massive 48.55% drop in Intersection-over-Union score. The consequences of learning from training data with different magnitudes and settings of label noise have been explored.

Spatial flood susceptibility mapping (FSM) is one of the key components of flood risk assessment. Recent studies showed the efficacy of hybridized machine learning (ML)-based models in predicting flood susceptibility. The quality of this prediction depends on the presence or absence of label noise in the training data collected from real world flood inundation maps. However, no study has been conducted to explore the effects of label noise in ML based spatial FSM. In the present study, we have explored the effects of four different random class label noises on the regression performances and model complexities of both standalone and hybridized ML based FSM. We also explored the optimal hyperparameter values of different hybridized ML and reported the robustness of each model in the presence and absence of label noise. The hybridized ML models showed greater efficacy in modeling spatial flood susceptibility in both cases.

The paper proposes a fuzzy logic based food recommendation with the concepts of BMI (Body Mass Index), age, recommended nutrients and income. In Bangladesh, most of the people are suffering from malnutrition as they have no clear idea about food nutrition and the case is worse in the rural area due to low income of people are living there. We have developed a fuzzy model that recommends addition or reduction of food items in daily food habit that meets nutritional needs and budget limits of rural people in the context of Bangladesh. Since people have variations in their food intake, we have focused mostly in their nutritional imbalances to find the appropriate food groups that can be suggested. Local food items have been checked and divided into low and high cost categories. By fuzzifying different parameters, a recommended food list is presented that will consider both the income and preference level of rural people.

The research proposes an approach that can automatically classify real world pictures of some traditional clothing worn in Bangladesh into the predefined classes using Convolutional Neural Networks (CNN). The research is driven by considering the growing market of online shops in mind. For classification purpose, we have collected clothing images from several online stores and labeled them accordingly. Our CNN model is based on the Google Inception model. For comparison purposes we have tried several architectures of CNN and some variations to see how our model perform against them. We have tested our own model with three different optimizers - SGD, Adam and RmsProp. Among these optimizers, RmsProp performed the best. The final result shows our model could classify the images of the training and testing set with 92.05% and 89.22% accuracy respectively.

Bangladesh suffers more for being susceptible to flood driven loses than relishing the natural advantages of being a riverine country. It is crucial for flood management department to identify the severity of flood and the factors behind it. Though three main contributors to flood occurrences are the Brahmaputra, Ganges and Meghna river basins, there are several factors such as rainfall, location, elevation and many more that also has prominent influences in causing flood in different regions of Bangladesh. This paper intends to identify the flood condition of different regions of Bangladesh by implementing different clustering techniques such as K-means, Expectation Maximization and Agglomerative Hierarchical clustering algorithm. This paper aims not only to analyze the factors that lead to flood condition but also observe the effectiveness of applying the chosen clustering methods for further analysis.

Bangladesh, a nation renowned for its rich fertile land and a population around 160 million, earns most of its living from agriculture. The nutrient rich lands help us providing year-round crop yields that play a crucial role for the economy of Bangladesh. Thus, this is important to deliberately work on agricultural planning and prediction models to ensure economic prosperity. The advancement of crop yields is significantly dependent on soil factors like Ph, nutrients and organic substances along with climatic factors like rainfall, temperature and humidity. Data of such factors are recorded to serve the purpose of scientific and statistical analysis. With the help of applying different data mining techniques on them, we are able to determine effective parameters to predict crop yield from different locations. This paper mainly focuses on the analysis to predict Bangladesh's four most yielding crops; wheat, jute, T-Aman and mustard. To carry out the whole experiment, we have analyzed soil properties of medium high land and high land from different sub districts of Bangladesh and also their respective climatic data and crop production of the last 6 years. For our analysis, we have applied different data mining techniques such as K-means, PAM, CLARA and DBSCAN for clustering and four linear regression methods to predict crop yields.

This study focuses on the temporally evolving geo-environmental characteristics of Dhaka. Dhaka is located at the center of Bangladesh. In the last few decades, it has undergone rapid development in terms of infrastructure and economy. The land use pattern in Dhaka is not the same as it had been previously. The focus has been to analyze the changes that have occurred in Dhaka, which consists of six smaller cities inside it. Using remote sensing techniques and various satellite images, the objects are classified into different categories. Multi spectral characteristics and spectral response patterns are analyzed and studied. Using the information extracted across different decades, changes in land use have been derived. Furthermore, changes in different features e.g. built up, crops, water bodies etc. are analyzed and studied across years to factor out the zones that are experiencing recent urban growth. This generalized pattern of urban growth can be a valuable finding for urban planning and policy for the government of Bangladesh.

The diagnosis of the different types of cancers has greatly increased among the population of Bangladesh over the last decade. From 2008-2010 the National Institute of Cancer Research and Hospital (NICRH) has confirmed 27281 cases of cancers among male and female patients. This research paper works to find the relationship of the diagnosis of such cancer patient in accordance with their occupation, educational status and tobacco smoking habits. Data was collected from the Cancer registry Report of 2008-2010 published by the Department of Epidemiology at NICRH. Using the dataset two ANFIS models were generated to evaluate the risk of the top five malignancies among male and female population of Bangladesh.

In this paper we have proposed a system that will be able to analyze the Optimum Crop Cultivation of Bangladesh based on the knowledge of Neuro-Fuzzy System (NFS). The Neuro-fuzzy system is the collection of two techniques: fuzzy logic and the neural network. The system can compute the yield of a certain crop by using the value of humidity, temperature and rainfall. By using this system farmer will be able to increase agricultural productivity. Hence, this will have an overwhelming impact on poverty alleviation, boosting employment rate, human resource development and food security. The dataset that we used to train our system was collected from the official website of Bangladesh Bureau of Statistics. It contains the humidity, temperature and rainfall values of thirty-three districts of Bangladesh that produced the most crops from the year 2007-2013. We considered a few major crops of Bangladesh, i.e., Rice (Aus, Amon, Boro), Wheat and Potato. By using this system, farmers can harvest maximum production of crops throughout the various seasons in the year.

The research proposes an approach for searching and sorting food list according to user preference. The system maintains a fuzzy database to store restaurants and their menus with some specific parameters for each item. System allows the user to specify food tastes that are of their interest, location name where they prefer to eat, approximated budget for each person and few other defined parameters as searching keyword. While searching, the system will fetch the entries from the database according to user defined parameters, convert each item's metadata to fuzzy parameters and pass the list to a fuzzy controller. Then the controller assigns a score as an output to each item. Finally the list is sorted in descending order. System allows the users to provide the feedback about the food which makes the system to be adaptive. System also gets scaled for each individual user as food taste varies significantly for individual users; hence the system provides more precise result and makes it a complete food searching engine.

In this paper, we are proposing a personalized music recommender service based on Mamdani Fuzzy Interference System (M-FIS). Collection of playlist is used for gathering users' choice and mood while listening to songs. Similarity between audio files is calculated based on Mel Frequency Cepstral Coefficients (MFCC). We have developed a recommender model based on M-FIS with the aforementioned similarities and playlists. We were able to gain an acceptable accuracy rate using FIS compared to other method reported in literature.

Virtual machine consolidation includes issues like choosing appropriate algorithm for VM selection for migration and placement of VMs to suitable hosts. The reason for migration is caused due to the overutilization or underutilization of physical hosts. Virtual machines are needed to be migrated from overutilized host to guarantee that the demand for computer resources and performance requirements are accomplished. Migration from underutilized host is needed to deactivate that host for saving power consumption. In order to solve the problem of energy and performance, efficient dynamic VM consolidation approach is introduced in literature. In this work, we have proposed multiple redesigned VM placement algorithms and introduced a technique to migrate VMs by clusters by taking into account both CPU utilization and allocated RAM. We ran our simulation on a cloud computing simulation toolkit known as CloudSim using PlanetLab workload data. We make an extensive performance analysis against the default VM placement technique that could be found in ColudSim architecture.

Big data analysis has a huge importance not only to students and researchers but also to small, medium and large businesses. This is mainly due to the fact that big data is almost abundant storage of critical information. Any kind of products and services that people enjoy or do not find very helpful, at least a handful of people around the world will always post about it on social media. In this paper, we have discussed a proper methodology for the utilization and interpretation of twitter data which lead to some interesting insights into the world of public opinions about the iPhone 6. We also included various feature popularities and male-and-female reactions to the product. We found mixed opinions, most of which were consistent with the general comments and opinions expressed by users.

Due to the fast growing computing need catered by Cloud infrastructure, energy consumption reduction has been an active field of research. Dynamic VM consolidation is one such research area where server load is monitored and intelligent decision is taken to optimize the usage of datacenters. In our research, we have designed an overload detection method and several VM selection methods for VM migration to achieve the trade off in VM consolidation. We have redesigned and updated an existing overload detection algorithm using heuristics and implement it in our new VM selection algorithms combining migration control and heuristics that was introduced in[9,10]. We have evaluated our algorithms through simulation on large-scale experiments driven by real-world workload traces from more than a thousand Planet lab VMs. From the comparison with existing energy aware VM consolidation methods, it is found that performance of proposed method incurs least energy consumption.

We want to predict the next song a user might prefer to listen based on their previous listening patterns, currently played songs and similar music based on music data. To calculate music similarity we used a Matlab toolbox that considers audio signals. We used association rule mining to find users' listening patterns and using the association rules we predict the next song the user might prefer. As we propose a music discovery service as well, we use the information of music listening pattern and music data similarity to recommend a new song.

Agricultural crop production depends on various factors such as biology, climate, economy and geography. Several factors have different impacts on agriculture, which can be quantified using appropriate statistical methodologies. Applying such methodologies and techniques on historical yield of crops, it is possible to obtain information or knowledge which can be helpful to farmers and government organizations for making better decisions and policies which lead to increased production. In this paper, our focus is on application of data mining techniques to extract knowledge from the agricultural data to estimate crop yield for major cereal crops in major districts of Bangladesh.

IaaS has become one of the most dominant features that cloud computing offers by providing access to computing resources in a virtualized environment. IaaS enables datacenter's hardware to get virtualized using virtualization, which allows Cloud providers to create multiple Virtual Machine (VMs) instances on a single physical machine, thus improving resource utilization and increasing the Return on Investment (ROI). The placement of VMs among hosts is interpreted as a bin packing problem in most of the time. In this work, we have implemented some of the bin packing solutions to address these VM placement problems. We ran our simulation on a cloud computing simulation toolkit known as CloudSim using Planet Lab workload data.

Air pollutants are really a hazardous problem in Bangladesh. This paper works on the relationship between the pollutants and the admittance of patients in the medical facilities and analyzes the reason behind the increase of the disease rate in the hospitals. The research collected medical data from the medical center named National Institute of Disease of the Chest and Hospital (NIDCH) that is located in Dhaka, Bangladesh and the air pollutants data of the city of Dhaka. This paper uses k-means clustering method for clustering different air pollutants in different seasons of Bangladesh. CART method is also used to classify the patients according to different rate of admission. The missing values of data were replaced by probabilistic mean median method.

Data Mining is the process of finding pattern or useful information from large volume of data. The goal of this paper is to find the reason behind the unusual high birth rate by applying data mining techniques, e.g., decision tree, neural network, Bayes Classifier, Ripper and Support Vector Machine. The datasets were collected from the baseline survey conducted by the maternal neonatal and child health programme by ICDDR,B. If we could find the reason(s), high birth rate at home could be avoided in future. Giving birth at home is very dangerous as many complications may arise during pregnancy as well during birth. From the opinions of experts and professionals, it could be said that the risk of mortality of new born during home birth is quite alarming and birth at hospital/clinics seemed to be the safest place to protect the health and well-being of the woman and her baby.

Handover is a very common event in Cellular Mobile network; however QoS could be severely affected by the handover performance. More handover cause more signaling traffic. Therefore, it is desired that handover should be done only when it is necessary. Besides, handover decision should be precise by taking account all possible options and considering the best one. For moving Mobile Station (MS) handover takes place more frequently. Some fuzzy logic based methodologies have already been proposed in literature to provide handover decision. In this paper, a method has been proposed to calculate speed and direction of MS relative to base station as a single metric using measurement data. Accordingly, a fuzzy logic based handover algorithm is implemented to avoid ping pong effect. By taking relative speed and direction, traffic load, signal strength and distance, the fuzzy inference system determines the best candidate neighbor based on the measurement reports from MS. Simulation has been carried out in Matlab environment and a comparison of different approaches has been performed. Simulation results demonstrate that proposed algorithm provides prediction based handover decision more accurately and avoid unnecessary handover and ping pong effect.

Cloud Computing helps to provide quality of service to the end users within required time frame. Serving user requests using distributed network of virtualized data centers is a challenging task as response time increases significantly without a proper load balancing strategy. There are many algorithms proposed in the literature to support load balancing in cloud environment. All of them have their merits and demerits. The inherent structure of load balancing is rather imprecise. In this research, we model the imprecise requirements of memory, bandwidth and disk space through the use of fuzzy logic. Then we design and evaluate an efficient dynamic fuzzy load balancing algorithm which could efficiently predict the virtual machine where the next job will be scheduled. We implement a cloud model in simulation environment and compare the result of our novel approach with existing techniques. Simulation results demonstrate that our fuzzy algorithm outperforms other scheduling algorithms with respect to response time, data center processing time, etc.

Replication in a data grid reduces access latency and bandwidth consumption. However, when different sites hold replicas of a particular file, there is a significant benefit realized by selecting the best replica from among them. The best replica is the one that optimizes the desired performance criterion such as absolute performance (i.e. speed), cost, security or transfer time. By selecting the best replica, the access latency can be minimized. We develop a predictive framework that uses data from various sources and predicts transfer times of the sites that host replicas. With this estimate, one site can request the replica from the site that has the lowest transfer time. We use a neural network (NN) for transfer time prediction of different sites that currently hold file replicas. We compare the results with a multi-regression model and the simulation results demonstrate that the neural network technique is capable of predicting transfer time more accurately than the regression based model.

Data replication is an excellent technique to move and cache data close to users. By replication, data access performance can be improved dramatically. One of the challenges in data replication is to select the candidate sites where replicas should be placed. We use a multi-objective model to address the replica placement problem. The multi-objective model considers the objectives of p-median and p-center models simultaneously to select the candidate sites that will host replicas. The objective of the p-median model is to find the locations of p possible candidate replication sites by optimizing total (or average) response time; where the p-center model finds p candidate sites by optimizing maximum response time. A grid environment is highly dynamic so user requests and network latency vary constantly. Therefore, candidate sites currently holding replicas may not be the best sites to fetch replica on subsequent requests. We propose a dynamic replica maintenance algorithm that re-allocates to new candidate sites if a performance metric degrades significantly over last K time periods. Simulation results demonstrate that the dynamic maintenance algorithm with multi-objective static placement decision performs best in dynamic environments like data grids.

We propose a static replica placement algorithm that places replicas to sites by optimizing average response time and a dynamic replica maintenance algorithm that re-allocates replicas to new candidate sites if a performance metric degrades significantly over last K time periods. Simulation results demonstrate that the dynamic maintenance algorithm with static placement decisions performs best in dynamic environments like data grids.

Summary form only given. Replication is a technique in data grid environment that helps to reduce access latency and network bandwidth. Replication also increases data availability and thereby enhances the reliability of the system. Selecting the best replica depends on several factors such as past behavior of the transfer, current state of the network as well as the state of disk device. Here, we develop a predictive framework with a neural network that uses the data from various sources and predicts transfer bandwidth. We compare our results with regression models and demonstrate that the neural network technique outperforms the regression model based predictors for large file transfers.