Thesis 2018

Browse

Search Results

Now showing 1 - 10 of 26

Food and Formalin Detector Using Machine Learning Approach
(East West University, 2018-09-17) Memi, Afsana Azad; Sultana, Nasrin; Tabassum, Kanij
Unethical use of formalin, in the preservation of food items posing threat to public health. Without chemical experts accurately Formalin detection is a time consuming and complicated task. Moreover, the presence of naturally occurring formalin in food items may interfere in detecting artificially added formalin. Purpose of the study was to develop a simple cost-effective and reliable detection technique that can detect contaminated food. Therefore, quantifying artificially added formalin and naturally formed extent it is important to dynamically detect food for the comparison. With this view in mind, we have applied different machine learning algorithms like Naïve Bayes, Logistic regression, Support Vector Machine, K-NN Classifier on fruit’s feature dataset to build a predictive model. We found that the K-NN algorithm works best in terms of accuracy. Finally using food conductance to electricity Rules have been developed and uploaded to the microcontroller unit. Combining with Arduino and the VOC HCHO gas sensor our own android application is able to detect 1-50 ppm of formalin. Several Tests are conducted and polynomial regression has been applied to predict the concentration of formalin in a given sample.
Facial Expression Recognition Using Subspace Learning On LBP
(East West University, 2018-05-05) asnem, Kazi Nuzhat T; Ahmed, Tazin
There is different types of methods that can recognize the facial expression but none of them were able to generate the accurate result due to the lack of generalizability. This field has a huge possibilities and can open new doors to human machine interaction. As a result the demand of recognizing the human expression correctly is increasing day by day. So there are many ways to recognize the facial expression. Here in this paper, we are trying to analyze the facial expression on different sub space. First we applied a conventional method, LBP. Then we tried to apply Principal Component Analysis (PCA). We tried another subspace algorithm called Kernel Principal Component Analysis. Then we compared the results. We compared the accuracy of recognizing facial expression of these two algorithm using BSVM tool.
QoS-aware Channel Allocation in Directional Wireless Sensor Networks
(East West University, 2018-09-22) Tithi, Sharmin Sulatana Sattar; Hasan, Md. Mehedy; Hoque, Monica
In WSNs, majority of the channel allocation mechanisms considered energy efficiency as the many objective and assumed data traffic with similar priority. However, the introduction of image and video sensors demands certain QoS from the channel allocation mechanism and the underlying network. Managing real-time data requires both energy efficacy and QoS assurance. In this paper, we first present a novel QoS aware channel allocation algorithm, QDCA (QoS-aware Channel Allocation in Directional Wireless Sensor Networks), that supports high data rate for real-time traffic. By exploiting node clustering and directional communication, the channel is allocated dynamically by mapping the data priority requirements of the transmitting to the appropriate channel. The proposed mechanism works in distributed manner to ensure bandwidth and end-to-end delay requirements of real-time data. At the same time, the throughput of non-real-time data is also maximized. We have also proposed an improved QDCA mechanism, IQDCA, that enhances data throughput by inducing little computational and communication overhead. Results evaluated in simulation shows that QDCA and IQDCA mechanisms exhibit enhanced performance in terms of average delay, throughput and fairness.
Analyzing Protein Structure and Exploring the Sequence of Protein using Machine Learning Approach
(East West University, 2018-04-21) Peu, Sharmin Sultana; Roy, Rajashree
Protein structure and sequence analysis is an important and essential problem. Now machine learning techniques have been widely used in bioinformatics. In this research we analyze the protein of structure and sequence and predict the class of protein sequence. Also find the accuracy of that class for different machine learning algorithm. For the data set we use the exploratory data analysis (EDA) and extracted 278866 protein features from the data set. We classify the features and measure the accuracy level of the three machine learning algorithm: Support Vector Machine (SVM), Naive Bayes and Random Forest (RF) approach for that protein sequence.
Verification of Feature Anomalies in Software Product Line Feature Models
(East West University, 2018-04-21) Saha, Birat; Rasel, Md. Fahim Shahrier; Shakib, Syed Nazmus
Products with new features need to be introduced on the market in a prompt step and organizations need to speed up their development process. Reuse has been suggested as asolution, but to achieve effective reuse within an organization a planned and preemptive effort must be used. Software Product lines are the most promising technique and it increases productivity and software quality and decreases time-to-market. In SPL, a feature model shows various types of features and seizures the relationships among them. As different configuration found in feature model it has high probability that some configuration is not correct. There could be anomalies in feature model which could lead invalid configuration every time. Number of rules proposed to identify those anomalies. Some tools can identities those anomalies. Verifying those rule with different tools ensure that those rules are correct and universal on every tools.
Detection of Brain Tumor Using Internet of Things
(East West University, 2018-05-10) Jahan, Ifrat; Rahman, Md. Lizur
The amount of brain tumor patients are increasing indescribably in the recent years and it has become a dangerous problem. For both men and women, brain tumor placed in 10th position of the leading cause of death. If it is possible to detect any disease before it started to damage, the chance of recovery from diseases gets the increase. Previous state-of-the-art techniques based on magnetic resonance images (MRI), provide fast and robust detection of tumor on the brain. But due to use of MRI images the complexity of these techniques become high. All these existing techniques classify the MRI images and detect the brain tumor, which is computationally expensive. We have proposed a stochastic method for automatic detection of brain tumor using Internet of Things (IoT). The proficiency of the method is scrumptious, because it measures the probability of brain tumor from our daily activities. In our experiment, we have used a portable wrist wearable device and two extra sensors, which can track our daily activities. Some common symptoms of brain tumor are used here to detect the brain tumor. We have proposed some equations, which can easily measure these symptoms from our daily activities. Experimental result for brain tumor patients‘ group and normal persons‘ group show the ability of the proposed technique in automatic detection of brain tumor. In this paper, we show the effectiveness of our method in automatic detection of brain tumor is demonstrated and it produces a better accuracy with the comparison to previous state-of-the-art brain tumor detection techniques.
Localization Of Mobile Submerged Sensor Using Cayley-Menger determinant and PSO Algorithm
(East West University, 2018-05-06) Lima, Mehebuba Naorin; Amin, Md. Faizul Ibne; Binte, Sumaiya
Underwater wireless sensor Networks(UWSNs) are usually deployed over a large sea area and the nodes are usually floating, due to their special environment. This results in a lower beacon node distribution density, a long time for localization, and more energy consumption. Currently most of the localization algorithm in this field do not pay enough consideration on the submerged mobile nodes/sensors. This paper investigates the problem of localizing submerged mobile sensors in a random direction having water current and provides a new mechanism to determine the coordinates of those sensors using only one buoy. In underwater wireless sensor networks (UWSN), the precise coordinate of the sensors that actuate or collect data is vital, as data without the knowledge of its actual origin has limited value. In this study, the method of determining the underwater distances between beacon and sensor nodes has been presented using combined radio and acoustic signals, which has better immunity from multipath fading. Cayley-Menger determinant is used to determine the coordinates of the beacon nodes. The velocity of this beacon nodes is determined using Particle swarm optimization. Finally here the unknown mobile nodes position is determined and updated using the velocities of beacon node/sensors and unknown mobile nodes which have found using PSO algorithm. Simulation results validate the proposed mathematical models by computing coordinates of mobile nodes with negligible errors.
A New String Matching Algorithm for Analyzing University Curriculum with Respect to Current Job Circular
(East West University, 2018-05-07) Al-Faruk, MD. Obaidullah; Hussain, K.M. Akib; Shahriar, MD. Adnan
Mining data from text is often becomes a crucial part of data mining tasks. With the growing tendency of using cloud and sharing more and more les over the internet, the necessity of applying a string matching algorithm in text mining has increased rapidly in present time. In recent years many pattern matching algorithms are proposed to enhance information retrieval from large le(s), especially in search engines as a mean of searching a certain term throughout multiple web pages to rank pages. These tasks require a faster string matching that can nd a certain pattern from a text with a very minimal waste of time. This can be ensured by using an algorithm that makes less character comparisons and pattern shifts while searching. In this paper, we're proposing a new algorithm named Back and Forth Matching (BFM) algorithm to perform string matching tasks in faster way by matching a pattern from both the forward and backward direction. A comparison of this algorithm with other algorithm shows a tremendous improvement in matching strings in large text les. For this advantage, we have implemented this algorithm in searching through universities course curriculum in Bangladesh context and compare it with the existing job circulars, in order to nd out how much the university curriculum is relevant with respect to current job markets. This will provide universities to learn about their laggings and thereby make necessary improvements as per the suggestions generated from our project.
Facial Expression Recognition Using Signed Local Directional Pattern
(East West University, 2018-04-22) Das, Nayan; Hasan, Kazi Md. Jamil
\Facial expression recognition has many implications nowadays. But due to lack of performance, human computer interaction is not a pleasurable experience yet. In this paper we have proposed a better version of the LDP (Local Directional Pattern). In the previous LDP calculation the Kirsch masks directional information were lost because of taking the absolute value. So we have proposed the sLDP approach. In this approach we are not going to lose the directional information of the Kirsch mask. Thats why we have taken the signed value for the LDP calculation. By applying sLDP we were able to get a signi cant improvement over the previous one for the 6-class expression of the cohn dataset. We also got a little bit of improvement of the 7-class expression as well. "
A New String Matching Algorithm for Analyzing University Curriculum with Respect to Current Job Circular
(East West University, 2018-05-07) Al-Faruk, MD. Obaidullah; Hussain, K.M. Akib; Shahriar, MD. Adnan
Mining data from text is often becomes a crucial part of data mining tasks. With the growing tendency of using cloud and sharing more and more les over the internet, the necessity of applying a string matching algorithm in text mining has increased rapidly in present time. In recent years many pattern matching algorithms are proposed to enhance information retrieval from large le(s), especially in search engines as a mean of searching a certain term throughout multiple web pages to rank pages. These tasks require a faster string matching that can nd a certain pattern from a text with a very minimal waste of time. This can be ensured by using an algorithm that makes less character comparisons and pattern shifts while searching. In this paper, we're proposing a new algorithm named Back and Forth Matching (BFM) algorithm to perform string matching tasks in faster way by matching a pattern from both the forward and backward direction. A comparison of this algorithm with other algorithm shows a tremendous improvement in matching strings in large text les. For this advantage, we have implemented this algorithm in searching through universities course curriculum in Bangladesh context and compare it with the existing job circulars, in order to nd out how much the university curriculum is relevant with respect to current job markets. This will provide universities to learn about their laggings and thereby make necessary improvements as per the suggestions generated from our project.

Thesis 2018

Browse

Filters

Settings

Sort By

Results per page

Search Results