Faculty of Engineering & Technology

Permanent URI for this community

https://dspace.nstu.ac.bd/handle/123456789/1

Browse

Now showing 1 - 20 of 78

A Comparative Analysis of Machine Learning Models for Liver Disease Prediction
(2024-05-13) Raihan Uddin
Liver disease is a widespread global health concern that demands early detection for effective treatment and improved patient outcomes. Traditional diagnostic methods have their limitations, necessitating further exploration into the use of machine learning to enhance predictive accuracy, accessibility, and cost-effectiveness in liver disease diagnosis. This study provides an in-depth analysis of machine learning classifiers such as Support Vector Machine, K-Nearest Neighbors, Decision Tree, Multilayer Perceptron, and Random Forest for liver disease prediction and compares their performance. The primary objective of this research is to identify the most effective machine-learning classifier for precise and accessible liver disease prediction. This research has the potential to significantly impact patient care, as it can aid in the development of efficient diagnostic tools, reduce the global burden of liver disease, and improve patient outcomes. This study's robust methodology includes feature selection via the Pearson Correlation matrix, leveraging the 10-fold cross-validation technique, and a balanced dataset generated using the Synthetic Minority Over-sampling Technique (SMOTE) on our collected dataset from distinct hospitals across Bangladesh. Our approach, employing a Random Forest classifier, achieved an impressive 80.5 percent accuracy in liver disease prediction. Furthermore, our model demonstrated a precision of 0.88 and a recall of 0.85, indicating its robust performance in identifying positive and negative cases. Keywords: Liver Disease, Pearson Correlation, Support Vector Machine, K-Nearest Neighbors, Decision Tree, Random Forest, Multilayer Perceptron.
A COMPARATIVE STUDY ON HEART DISEASE PREDICTION USING DIFFERENT MACHINE LEARNING CLASSIFICATION ALGORITHMS
(2020-01-30) MD. IMAM HOSSAIN & MEHADI HASAN MARUF
Heart disease is the leading cause of death over the past 10 years all over the world. The risk factors of heart disease may hold numerous measures and considering all the measures increases the time of medical practitioners for the decision-making process. The main aim of this work is to pave the way to maintain the huge datasets with the appropriate measures and choose the right approach which would provide a faster and more accurate prediction of heart disease depending on the accuracy rate of the measures. In this thesis, we examine and compare the performance metrics of different machine learning classification algorithms to predict heart disease. This comparison shows the different accuracy, precision, and recall rates of different techniques and the reasons behind their variations. We used collected data that we collect from the different hospitals in Bangladesh. We preprocess our collected data and then divide it into two sections named training and testing datasets. The Logistic Regression, Naive Bayes Classifier, K-Nearest Neighbor, Support Vector Machine, Decision Tree, Random Forest, and Multi-Layer Perceptron Neural Networks techniques have been investigated in this research. By the end of the implementation part, we have found Random Forest is giving the maximum score in all sectors like accuracy, precision, recall, and f1 score in our dataset, and Multi-Layer Perceptron is performing very poorly. Random Forest gives a maximum of 90 percent accuracy and KNN gives the second-best 85 percent accuracy. Other algorithms like KNN, SVM, and Decision Tree also show overall good performances. The reasons for variations of these different techniques by analyzing their characteristics and behavior with respect to the dataset have been understood by the study conducted for this thesis. The research outcome is the implementation of machine learning on healthcare data and finding a better technique for prediction.
A DIGITAL SYSTEM FOR ATTENDANCE, PAYMENT AND PORTABLE ADMIT CARD GENERATOR
(2022-11-30) MD. AKRAM
Digital attendance system takes attendance and calculates the average attendance percentage for a certain period through digitally. The result of attendance system is used in payment system to identify whether a student is able to attend in exam and download admit after completing the examination fee. This system can minimalize the time, effort, and cost of manual attendance system. The whole system has three modules i.e. Admin module, Teacher module, and Student module. Everyone has their own identity based on their rule. Student can open their login credential using basic information and enroll in courses, monitoring attendance on a regular basis, complete course registration and exam fee payment through MFS or Internet Banking and download admit. Teachers’ credentials are used for login in the system to take attendance. Teacher has the authority to enroll in a specific course. The whole system has one administrative credential that is used for authenticate the others necessary information and modify access rights. We used HTML, CSS and JavaScript for front end and PHP and Laravel framework used for back end design. We are trying to keep the entire project easy, efficient and convenient and designed in such way that specific tasks can be completed in less time. The essential time required for taking attendance to calculating average attendance, paying exam fee for acquiring admit to seat in exam must be decrease if the system implied in academic purpose. If this project fully implemented than it will save time, effort and money. Keywords: Digital Attendance System, Digital Payment, Portable Admit Card, Examination.
A Hybrid Deep Learning Approach for Network Intrusion Detection and Prevention
(2025-07-30) Niloy Kumar Joy
With the ever-changing online threats that are entailing fast development, there are increased concerns on strong and intelligent network security. The static rule-based traditional intrusion detection systems (IDS) based on signature-based approach are not able to detect new or advanced attacks. The present thesis suggests a combination of both deep learning technologies based on Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM) networks to improve the detection of intrusion and allow the intrusion prevention proactive measures. By combining the spatial feature extraction capability of CNN and the temporal dependence capture capacity of LSTM, CNN-LSTM model can be said to make best use of the two models when the domain is network traffic. On the NSL-KDD dataset which is a benchmark dataset that comprises of different classes of attacks namely DoS, Probe, R2L, and U2R, the model was trained and tested versus other deep learning models like the MLP, standalone LSTM and autoencoder. The presented model beat others, with the highest accuracy of 95.79 percent, and better precision, recall, and F1-score. Besides detection, this paper discusses rudimentary prevention systems to narrow the response time and constrain attacks. The findings prove that a hybrid deep learning architecture is more likely to offer a more precise, flexible, and integrated answer to intrusion detection and prevention. The study can be used to develop intelligent cyber security solutions and result as the precursor of real-time and self-learning IDS applications in the future. Keywords: Network Intrusion Detection, CNN-LSTM, Deep Learning, NSL-KDD Dataset, Cybersecurity, Intrusion Prevention, Hybrid Neural Network, Anomaly Detection
A Hybrid GA-PSO Optimized Machine Learning Framework with Strategic Data Augmentation for Early Autism Spectrum Disorder Detection and Risk Factor Analysis
(2023-06-30) Argina Akter
This research presents a robust machine learning framework for the early detection of Autism Spectrum Disorder and the analysis of its associated risk factors. The study leverages a comprehensive dataset of 288 individuals, encompassing behavioral, developmental, familial, and socio-demographic attributes. Five machine learning classifiers—Support Vector Machine, Decision Tree, Random Forest, Multilayer Perceptron, and Logistic Regression—were implemented and enhanced through a novel two-pronged approach. First, a true hybrid Genetic Algorithm-Particle Swarm Optimization technique was employed for superior hyper-parameter tuning, significantly boosting model performance. Second, strategic data augmentation was performed using Conditional Tabular Generative Adversarial Networks combined with boundary sampling to effectively mitigate data scarcity. The optimized Random Forest model achieved exceptional performance, with an accuracy of 98.85% and an Area Under the Receiver Operating Characteristic Curve of 0.9989. The synergistic combination of GA-PSO optimization and data augmentation proved transformative, enabling multiple models to achieve perfect classification metrics of 100% accuracy on augmented datasets. Model robustness was rigorously validated through stability analysis across multiple data splits (70:30, 75:25, 80:20) and, critically, on a completely unseen test set, where the best model maintained a high accuracy of 94.1%. Beyond predictive performance, a sensitivity analysis was conducted to identify and rank the most influential risk factors. "Limited Speech" emerged as the most critical predictor, followed by parental age at childbirth (i.e., "Mother age at childbirth" and "Father age at childbirth") and key developmental milestones such as "Ability to walk" and "Ability to respond when called by name". This study conclusively demonstrates that the integration of hybrid metaheuristic optimization and strategic data augmentation creates a highly accurate, reliable, and generalizable tool for early ASD screening. The framework provides a practical solution for overcoming data limitations in clinical settings and offers data-driven insights into key risk factors, thereby supporting earlier intervention and informed clinical decision-making. Autism Spectrum Disorder (ASD) is a complex neurodevelopmental condition characterized by social interaction challenges, restricted interests, and repetitive behaviors. Early detection is critical for timely intervention, yet conventional diagnostic methods are often subjective, time consuming, and inaccessible in low-resource settings. This research develops a robust machine learning-based framework for ASD detection using a multidimensional dataset of 288 individuals aged 3–27 years, encompassing behavioral, developmental, health, familial, and socio demographic attributes. Five classifiers—Support Vector Machine, Decision Tree, Random Forest, Multilayer Perceptron, and Logistic Regression—were implemented and further enhanced through hybrid Genetic Algorithm–Particle Swarm Optimization (GA-PSO) for hyper-parameter tuning. To overcome data scarcity, Conditional Tabular GAN (CTGAN) with boundary sampling was employed for synthetic data augmentation. The optimized and augmented models achieved significant performance improvements, with Random Forest enhanced by GA-PSO yielding near perfect accuracy (>98%, AUROC ≈ 0.999). Rigorous validation, including stability analysis across multiple data splits and testing on a completely unseen dataset, confirmed the model’s reliability and generalizability. Sensitivity analysis identified and ranked key risk factors, with limited speech, parental age, and developmental milestones emerging as dominant predictors. This integrated framework demonstrates the potential of hybrid optimization and augmentation strategies to deliver an accurate, scalable, and clinically meaningful screening tool for ASD, bridging the gap between machine learning innovation and real-world healthcare needs.
A STUDY ON THE IMPACT OF INTERNET USAGE ON THE ACADEMIC PERFORMANCE OF THE STUDENTS
(2021-10-29) SHAJID HOSSAIN HEMAL
Education systems have greatly changed with the emergence of Internet. It has a significant impact on how students learn things and how educational institutes operate. But its impact can also be contradicting. Internet addiction can slowly poison the minds of our youths and stand in the way of pursuing their goals. Proper analysis of the effects this technology has on today’s youth can help determine the factors that allow a student to succeed academically and those that do not. Also, predicting the academic performance beforehand can help determine the changes that must be incorporated to improve the educational system. So, this research attempts to accurately analyze the effects of internet usage on student’s academic progress and also predict the class performance of the students by using different machine learning algorithms. About 700 student’s data were collected for this study from an offline survey taken from different schools, colleges and universities, all situated in Noakhali district. These data will be analyzed with the help of statistical evaluation methods and machine learning algorithms such as Logistic Regression, Decision Trees, Random Forests and Naive Bayes Classifier. The results from these models will be analyzed using different performance metrics to determine the factors that contribute the most in their academic activities.
AN EFFECTIVE APPROACH FOR EARLY LIVER DISEASE PREDICTION
(2020-12-08) FARIA AFRIN
Liver is one of the main organ of our body. It will be functioning normally even when it is partially damaged, therefore problems with liver patients cannot easily discovered in an early stage. Patient’s survival rate can be increased by an early identification of liver problems. Liver disease can be diagnosed by observing the levels of enzymes in the blood. Many researchers working on this issue and try to find the best algorithm with using Liver patient dataset which is suitable for predicting disease in early age. In this research liver patient dataset is investigated for building different classification models to get better result by comparing with the accuracy of the classifiers for disease prediction. Real data in this purpose is collected from the hospitals. This paper uses different classification methods including Bagged Trees, Support vector machine (SVM), K-Nearest Neighbor (KNN), Fine Tree classification methods. It can give a good impact on the liver disease diagnosis and it can be beneficial for physicians and can help to reduce the cost of diagnosis in the medical sector.
AN EFFECTIVE TECHNIQUE ON EARLY HEART DISEASE PREDICTION
(2020-12-10) TANZINA RAHMAN HERA
Heart disease is the most threatened issue in a human body. Most of the people in the world are affected by heart disease which increases the death rate of humans considerably. The detection of heart disease is the most trivial task for medical researchers which cannot be done more accurately. The early detection, manual prediction and prevention is a complex task and troublesome. So there is an urgent need for a well-designed method for the early detection of heart disease. Thus it is required to implement the automated system by considering the recent computer technologies which can help medical researchers to diagnosis heart disease firstly and accurately. Now a day’s artificial neural network has been widely used as a tool for solving many decision modelling problems. This research enlightened a number of structures in Artificial Neural Network by varying the configurations of neural network algorithms like Radial Basis Function Neural Network, Recurrent Neural Network, Decision tree classifier and Support Vector Machine Neural Network. This research also compared those Neural Network models on the early detection of heart disease and chose the best neural network model among them to detect the heart disease early. The Network Structure that showed more accuracy and efficiency is chosen among them.
An integrated mobile app for rice disease detection and weather prediction
(2024-05-30) Fazla Rabbi
This project seeks to create a cutting-edge mobile application that combines the capacity for disease detection in rice crops and precise weather forecasting for agricultural areas. Rice is a staple crop for billions of people around the world, making agriculture a crucial economic sector. However, if these diseases are not promptly detected and treated, rice crops are vulnerable to a number of illnesses that can result in significant output losses. The success and growth of crops are significantly influenced by the weather as well. In order to provide rice farmers and other agricultural stakeholders with a complete solution, the suggested mobile app makes use of cutting-edge technology, such as machine learning and the integration of real-time weather data. The project's primary goals are as follows: 1. Rice Disease Detection: In order to recognize common illnesses impacting rice crops, such as blast, bacterial leaf blight, and sheath blight, the app will use picture recognition and machine learning algorithms. With the use of an app, users may take pictures of damaged rice plants using their mobile devices, and the program will instantly analyze and diagnose the situation, enabling farmers to take quick corrective action. 2. Weather Prediction: To deliver precise weather forecasts for particular agricultural areas, the app will link existing sources of weather data and apply cutting-edge predictive models. Farmers will be able to make informed decisions about planting, harvesting, and pest control thanks to this feature, which will contain data on temperature, humidity, precipitation, wind speed, and other pertinent elements. 3. Crop Management Insights: Based on disease detection and meteorological information, the app will provide actionable insights. Users will get advice on crop management techniques, such as when to apply pesticides, when to schedule irrigation, and how to rotate their crops. Losses will be reduced and crop yields will be optimized as a result. 4. User-Friendly Interface: Farmers with various degrees of technological ability will be able to utilize the mobile app because of its user-friendly layout. To reach a large user base, it will be accessible on both the Android and iOS platforms. 5. Data Privacy and Security: The app will use strong encryption and abide by data protection laws to guarantee the security and privacy of user data. This project intends to equip rice farmers with the resources and knowledge required to improve crop yield and sustainability by combining disease detection and weather forecasting in a single mobile application. Increased yields, a decrease in the use of pesticides, better resource management, and improved lives for rice farmers are all potential effects of the app. Keywords: Rice disease detection, mobile app, weather prediction, agriculture, machine learning, image recognition, crop management, agricultural technology, data privacy, sustainability.
ANALYSIS OF MOBILE PHONE RADIATION EFFECT ON HUMAN BODY USING SPECIFIC ABSORPTION RATE20
(2021-03-30) SAGAR MAHMUD
The use of mobile phones is rising everyday both in civic and rustic areas in Bangladesh. Every mobile phone disperses electromagnetic energy. This radiation or electromagnetic wave can penetrate into the human body. Specific absorption rate and change in temperature are the vital parameter to find the dominant effects on the human body. Many researchers have done a lot of work on the effects of radiation for his or her country. But in Bangladesh, no visible work has been done yet. The importance of the health hazards due to radiation from the mobile phone has not been understood among the common people. Because it cannot be seen, felt or smelt and it is very hard to realize the evil part. Importance of mobile phone cannot be ignored because it’s helps to keep in touch with family, friends and others to perform various tasks. This paper explores the radiation impact of mobile phones. Mathematical equations are used to evaluate the specific absorption rate and change in temperature at skin, fat, bone, brain, eye, muscle tissues at the frequency of 800MHz, 900 MHz, 1800 MHz, 2100 MHz, 2.5GHz, 2.6GHz, 3.4GHz, 3.5GHz, and 3.6GHz respectively. The calculations are performed at different distances and exposure times of mobile phone. A simple radio frequency detection circuit has been designed to find the radiated power of different frequency. The highest specific absorption rate is calculated for skin, brain, eye tissue. Child tissue absorption rate is higher than the adult. When the radiated power of mobile phone is high the specific absorption rate for different tissue crosses the safety limit. This paper also recommends the ways to diminish the effects of SAR. It is suggested that these methods will decrease the health risks. This work will be very helpful to understand the bad effect of using mobile phones and the way of reducing this effect.
ANALYSIS ON IMPACT OF PRECEDING EDUCATIONAL PERFORMANCE ON THE RESULTS ACQUIRED IN HIGHER STUDY
(2020-12-08) MONIR HOSSAIN
Academic performance has been one of the most important goals of the educational process. Academic achievement is a key mechanism through which adolescents learn about their talents, abilities and competencies which are an important part of developing career aspiration. It is also a major goal, which every individual is expected to perform in all cultures. Many researches have been conducted to investigate factors affecting students’ performance in university and colleges. The aim of this research is that determine the impact of students academic result in school and college on student’s performance in university level. We determine the effect of SSC and HSC result in honor’s result. We find out the importance of various subject including Mathematics, Chemistry, Physics and Biology by determine individual impact on honors result in department of engineering and biology faculty. We also analysis the university honor’s result categorized by gender, department, SSC board and university admission merit. The research is based on student records based on the dataset collected through assessment from university databases. Data analysis technique- statistical analysis technique was used to extract useful information from the database using Rapid Miner software. The results obtained by using various charts like scatter, multiple scatter and bar charts. This analysis result helped to find out many useful information based on which proper guidance and suggestions are designed for students. Students will know more that which subject should be given more importance. Students will also see examples of good results in university without getting good results in the past and they will be more energetic to good results.
ARTIFICIAL INTELLIGENCE AND NEURAL NETWORK BASED MATERNAL AND FETAL HEALTH RISK LEVEL PREDICTION AND SENSITIVITY ANALYSIS DURING PREGNANCY
(2022-11-30) Zarin Tanzim
Health of women throughout pregnancy, childbirth, and the postpartum period is referred to as maternal health. To ensure that women and their unborn children achieve their maximum potential for health and wellbeing, each stage should be enjoyable. The majority of pregnancies and deliveries are successful, but complications do happen occasionally and when they do, they can have devastating effects on both mothers and infants. Through a better understanding of risk factors, increased surveillance, and more early and suitable interventions, predictive modeling has the potential to enhance outcomes and assist gynecologists in providing better care. Dataset used for analysis is collected from the local hospitals of Noakhali District. For the analysis, the main risk factors considered are age, Blood pressure (systolic), Blood pressure (diastolic), Body temperature, maternal heart rate, blood glucose, hepatitis B, TSH, serum SGPT, serum uric acid, fetal heart rate, amount of amniotic fluid ,fetal movement and obesity. The dataset contains the details of these features of women during their pregnancy. Data preprocessing is done by mapping naming value, handling missing value by prediction, selecting important feature and feature scaling etc. Correlation is also checked for numerical values. Data is then divided for training and testing, and the best accuracy for each prediction model is then determined. These models include Multinomial Logistic Regression, Naive Bayes, K-Nearest-Neighbors, Support Vector Machine (SVM), Decision Tree, Random Forest, and Neural Network Multilayer Perceptron. Accuracy of MLR, MNB, KNN, SVM, DT, RF and MLP algorithm is 89.58%, 80.13% , 100.0%, 90.23%, 100.0%, 100.0% and 100.0% respectively. The DT, RF and MLP algorithms also gives the best precision, recall, f1-score and ROC_AUC score i.e. all are 100.0%. In comparison to other algorithms, the DT, RF, and MLP perform best in terms of accuracy, precision, recall, f1-score, and ROC_AUC score. The input and target variables will also be subjected to sensitivity analysis to determine which input parameters have the greatest impact on risk. According to sensitivity analysis, the following factors have a significant impact on the risk to the mother's and the fetus's health: blood pressure(systolic & diastolic) fetal movement, fetal heart rate, and the amount of amniotic fluid. By avoiding maternal and child morbidity during pregnancy for areas of Bangladesh, this effort will help to improve healthcare for expectant mothers and their fetuses.
Automated Class Routine and Exam Schedule Management System(ACRESMS)
(2020-12-08) SHABBIR AHAMMED
Every educational institution needs class routine and exam schedule for its activities. If making this class routine and exam schedule will generate automatically, it will be very amazing. Automated Class Routine and Exam Schedule Management System(ACRESMS) is very important in any educational institutions. The authority can create class routine and exam schedule by clicking this system, generate class routine, update routine, generate exam schedule, update exam schedule and deleted it easily. This system saves time and reduce paper based work. We will divide the system into different segments such as teacher’s information, course information, classroom details, teacher details, time management, session, years, teacher summery etc. Every segments have contained different database for clear concept. We use here MySQL database to implement the system. The purpose of the project is to automated making ACRESMS which is user friendly, simple and fast. This system will be helpful for any institutions because there are few paper based task, student can get class schedules and exam schedule easily and admin can handle this system without any problem because it will be online based system. This system reduces complexity of class room, teacher availability problem, examination hall room, invigilator availability and paper based complexity. This system will be user friendly system.
Automated Surface Roughness Prediction from Scanning Electron Microscope Images using Convolutional Neural Network
(2024-05-13) Md. Shariar Kabir Asif
In recent years, the advent of Convolutional Neural Networks (CNNs) has opened up new avenues for advancing Surface Roughness (SR) prediction methodologies, particularly through the analysis of Scanning Electron Microscope (SEM) images. However, notable gaps existed in the literature regarding the application of CNNs to SEM images for SR prediction. This research addresses these existing gaps by employing CNN to analyze SEM images for SR prediction, particularly focusing on the comparative analysis of different magnification levels. Three distinct datasets, magnified at 150X, 250X, and 500X, were utilized, comprising 2097, 2103, and 2102 images respectively. These images undergo preprocessing techniques to enhance the CNN model's ability to generalize to new images. Subsequently, a sequential CNN model, comprising 27 layers including convolutional, max pooling, batch normalization, flatten, and fully connected dense layers, is developed and trained on the datasets. The study provides detailed comparative analyses of accuracy, precision, recall, and F1-score across magnification levels. Results indicate that the dataset magnified at 500X consistently outperforms the others, exhibiting superior accuracy (75.7%), precision (0.65), recall (0.72), and F1-score (0.72). This suggests that higher magnification levels provide finer details and clearer images, enabling the model to discern subtle features with increased accuracy. Additionally, the 500X dataset exhibits a better balance between minimizing false positives and false negatives, making it more suitable for real-world applications requiring detailed analysis of microscopic structures. These findings underscore the importance of selecting appropriate magnification levels in SEM imaging for accurate SR prediction. Keywords: Surface Roughness; Surface Roughness Prediction; Scanning Electron Microscope (SEM); Convolutional Neural Network (CNN); Magnified SEM Images
Bangla Handwritten Character & Digit Recognition Using Convolutional Neural Network
(2021-10-29) TANVIR HOSSAIN
Bangla handwriting recognition is an important and quite popular for the last few years. This paper proposes a Convolutional Neural Network(CNN) model to recognize the Bangla Handwritten Characters and Digits which contains 50 Character Classes and 10 Digit Classes. The most challenging part is to reduce the noise because of the different shapes of Banlga characters and digits. We propose to use a reliable and large dataset to get the best accuracy output with our proposed Convolutional Neural Network(CNN) model. We also compare the accuracy with the existing model. Finally, we hope we will make a model with less error and the best output result.
BLOCKCHAIN TECHNOLOGY FOR IOT SECURITY FOR A SMART HOME
(2018-12-30) MOHONA MOJUMDER
One of the key challenges of IOT world is security. In this article, my attempt is to critically view the use of blockchain technology to secure IOT. Blockchain is used for smart home so that the security can be enhanced. Blockchain enriches the security level.In this technology,the existing database can only be expanded and the previous record can not be changed.Local blockchain is a secure and private blockchain that is mined and stored by one device which is always in online. The local blockchain is only maintained by it’s owner.To increase the security level and to secure the data from sharing external environment, the blockchain technology is used.
CLASS ROUTINE DESIGN SYSTEM AND EXAM HALL INVIGILATION MANAGEMENT
(2021-03-30) RATUL PROSAD
This Project name is “Class Routine Design System and Exam Hall Invigilation Management”. This is a web based application. In Class Routine Design System mainly, focusses on creating automatic class routines with teacher’s requirements. Most universities handle this allocation process with a manual procedure. The manual procedure like this gives way to various challenges and is inclined to mistakes. A better approach to reliably schedule class routine is to utilize a computer assisted web-based system. This project has mainly two parts or two panel, Admin panel & User panel. In admin panel we get some information like courses information, teacher’s information, room’s information etc. We can Update, Delete & Add this information’s. There have teacher’s requirement for creating class routine and semester wise course offer distribution for teachers. We can update and view all this field in admin panel. After that we made a class routine using all of this field. In another panel is user panel. In this part we get all of information about course, teacher’s and rooms. In this panel we can see all this information & routine. Update, add and delete option in this panel also included. Both of this panel, we get routine separate by days, teacher’s and semester wise. We create this application for genetic algorithm, python. We create this application for reduce the complexity of routine creation manually, to reduce time to solve a problem and to reduce routine version. In Exam Hall Invigilation Management, proposes an improved algorithm to achieve automatic examination arrangement for invigilator based on greedy method, the algorithm can support the priorities better, set maximum frequency of invigilating and satisfy other personalization examination arrangement requirements, and also it can configure to allocate any numbers of invigilators in different examination halls in such a way that each invigilator will get equal amount of duties. This algorithm has written and implemented in java script language. Keywords: Routine Allocation System, Web Based Application, Invigilator, Examination.
CLASSIFICATION OF EARLY AND LATE BLIGHT DISEASE OF POTATO USING CONVOLUTIONAL NEURAL NETWORK
(2020-12-08) Jesmin Akther
Crop diseases are a major threat to cultivate, and need to supervise growth and detrimental diseases in time. Potato early and late blight disease symptoms and detection both are vague to distinguish and isolate. The rising combination of smartphone penetration and recent advances in deep learning has paved the way for smart device assisted disease prognosis. Relying on pure naked-eye observation to detect and classify diseases can be very cumbersome. Absolute detection process proves to be effective and convenient for researchers. The technique of training deep learning models on increasingly vast and globally available image datasets presents a clear path toward smartphone aid crop disease experiment around the world. The color and analyze layer features are used to best match to recognize and classify different agriculture produce into early blight and late blight affected disease. Both features prove to be very effective in disease detection. This paper deployed a Sequential convolutional neural network (CNN) model to detect and identify diseases in real time survey potato leaves labeling early and late blight. In addition to normalization, divide and extract the images to prepare data prior CNN. This work adopts slight variation during CNN model finalization with the help of Tensorboard analysis. This analyzed effectively optimized model validation accuracy each layer by layer hierarchically. Best layers ensure to create the final model which assures 94% accuracy for this dataset and short timing classification. The experimental results indicate that the approach significantly can be modified model accuracy in automatic detection of both affected bight.
Climate Forecasting in Bangladesh Using Distinct Artificial Intelligence Techniques
(2024-10-02) Jannatul Naim
In Bangladesh, a nation that is primarily dependent on agriculture and is extremely vulnerable to weather-related dangers, weather forecasting is of utmost importance. This research paper aims to develop accurate weather prediction models for rainfall, flood, temperature, and drought using distinct artificial intelligence regression techniques, including Support Vector Regression (SVR), Decision Tree, Random Forest, k-Nearest Neighbours (KNN), and Multilayer Perceptron Neural Network (MLPNN). Additionally, the optimization algorithms Genetic Algorithms (GA) and Particle Swarm Optimization (PSO) are sequentially applied to enhance the performance of these models. We utilized historical weather data from 1980 to 2022 provided by the Bangladesh Meteorological Department (BMD). Splitting the data into 80% for training and 20% for testing, we evaluated model accuracy using Root Mean Square Error (RMSE) and Mean Squared Error (MSE). Among the techniques tested, Random Forest showed the best performance, demonstrating its efficacy in weather prediction. Therefore, we will use Random Forest to predict the temperature and rainfall for the next 10 years (2023 - 2033). From our research, we have found that over the next 10 years, the average temperature in the country will increase by approximately 0.5 degrees Celsius, and the summer season will be longer. The winter season will be shorter, and the average temperature during winter will increase compared to previous years. The annual average rainfall in the country will decrease, and this reduced rainfall, coupled with higher temperatures, will nearly double the risk of drought. Additionally, due to climate change, the frequency of floods is expected to increase suddenly in the coming years, and there is a possibility of over flood in some years. The results of this study will advance the field of weather forecasting in Bangladesh by giving insight into the effectiveness of various AI regression approaches and optimization algorithms. Accurate weather forecasting can have a substantial positive impact on a number of industries, including agriculture, disaster relief, and public safety. Keywords - Weather Prediction, Artificial Intelligence, Regression Techniques, Support Vector Regression (SVR), Decision Tree, Random Forest, k-Nearest Neighbours (KNN), Multilayer Perceptron Neural Network (MLPNN), Optimization Algorithms, Genetic Algorithms (GA) and Particle Swarm Optimization (PSO), Weather Data.
Combating Deepfake News in Bengali: Detection and Prevention of Manipulated Social Media’s Content Using Machine Learning
(2024-03-30) MD. JANE ALAM
With the rapid growth of digital media and social networks, the spread of misinformation and manipulated content has become a significant concern. Nowadays, misinformation and false rumors mostly come from social media platforms. Deepfake technology, which allows the creation of realistic but fabricated news, audio, images, and videos, has emerged as a potent tool for the production and dissemination of deceptive content. While the detrimental impacts of deepfake news are widely recognized, research efforts in combating this phenomenon are largely focused on English and a few other major languages. On the other hand, Bengali grammar and language are significantly more difficult and essential to learn. This research paper aims to address this gap by proposing a comprehensive approach for detecting and preventing deepfake news specifically in the Bengali language. This study utilizes a newly released set of Bangla fake news dataset, marked by experts. It then employs Bengali-based embeddings for machine learning classifiers and utilizes trained bidirectional encoder representations from transformers (BERT) in Bengali to determine sentiments in Bengali grammar. The results show that increasing the training data consistently enhanced the BERT-based classifiers' performance, surpassing that of ML classifiers. When compared to prior studies on detecting Bengali fake news, the current study's findings indicate a more significant improvement. Keyword: Bangla Fake News, BERT classifier, Fake News Detection, Bangla

Browse

Browsing Faculty of Engineering & Technology by Title

Results Per Page

Sort Options