MS

Permanent URI for this community

https://escholar.umt.edu.pk/handle/123456789/14746

Browse

Now showing 1 - 20 of 254

4mc-rf
(UMT Lahore, 2020) FAJAR ARSHAD
N4-methylcytosine 4mC is an essential epigenetic modification that occurs enzymatically by DNA methyltransferase. 4mC sites exist in prokaryotes and play a vital role in regulating gene expression, DNA replication, and cell cycle. The efficient and accurate prediction of 4mC sites has a significant role in the insight of 4mC biological properties and functions. Therefore, we have proposed a sequence-based predictor, namely 4mC-RF, for identifying 4mC sites in prokaryotes by integrating statistical moments along with position and composition dependent features. Relative and absolute position based features are computed to extract the optimal features. A popular machine learning classifier Random Forest was used to training the model. Validation results were obtained under rigorous processes of Self-consistency, 10-fold crossvalidation, Independent testing, and Jackknife testing giving 95.01%, 95.02%, 97.02%, and 95.36% accuracies. Our proposed model depicts the highest prediction accuracies as compared with the literature results. Thus, the developed 4mC-RF model was constructed into a web server. A significant and more accurate predictor of 4mC Methylcytosine sites helps experimental scientists gather results moderately.
A blockchain-based framework to make rice crop supply chain transparent and reliable in agriculture
(UMT Lahore, 2023) IBTESAM UR REHMAN
Rice is one of the major food crops across the globe and its quality as well as safety ishighly associated with human health. It is widely used in making different by-products including rice flour, rice bread, noodle, rice vinegar, and other products. Therefore, the rice supply chain has grabbed increasing attention due to the high demand for food safety research. Furthermore, the malpractices in the rice supply chain influence the motivation of farmers by generating low revenue after putting high efforts into cultivation. In addition, the government also suffers huge economic losses by paying heavy amounts to import rice crops in other countries during the off-season. These issues occur due to the lack availability of reliability, trust, transparency, traceability, and security in the rice supply chain. In this research, we have proposed a secure, trusted, reliable, and transparent framework based on Blockchain for rice crop supply chain traceability from farm to fork. A new crypto token Rice Coin (RC) has been introduced in order to keep a record of all transactions between the stakeholders of the rice supply chain. Moreover, the proposed framework covers the economic model, and crypto wallet and introduced an Initial coin offering (ICO) for RC. Based on smart contracts a transaction processing system has been developed for transparency and traceability of rice crops as well as conversion of RC to fiat. Moreover, Inter Planetary File System (IPFS) has also been introduced in this research to store encrypted data of companies, retailers, and farmers to increase security, transparency, and availability. In the end, experimental results indicate that the proposed framework has better performance than already available supply chain solutions in terms of transaction verification time, transaction average gas cost, and new block latency
A blockchain-based framework to make sugarcane crop in agricultural supply chain transparent and reliable
(UMT Lahore, 2022) Muhammad Asad Khan
Sugarcane crop is used for making several ecofriendly and food products making it one of most importantbcrop in world. Sugarcane crop contributes in 60% sugar production of world. The malpractice in supply chain effects farmers’ motivation with low profits even after putting a lot of efforts in cultivation and governments bear economic losses by paying extra money to import sugar from other countries. Sugarcane supply chain processes are not transparent due to which it is difficult for authorities to keep track of malpractices and illegal exports. This paper proposed a blockchain-based framework for sugarcane supply chain traceability and reliability. A new token Sugar Coin (SC) has been introduced to keep track of transactions between supply chain entities. Furthermore, our proposed framework also covers crypto wallets, economic model and Initial Coin Offering (ICO) of SugarCoin. Additionally, a complete transaction processing system have been proposed using smart contracts for traceability of sugarcane trading and conversion of SC to fiat and vice versa. The Inter Planetary File System (IPFS) have been introduced for storing encrypted private data of farmers, companies and retailers to enhance availability, security and transparency. Finally, experiment results depict that proposed framework approximately performs better than already existing supply chain projects in terms of new blocks latency, transaction per minute, transactions average gas fee and times for transaction verifications.
A blockchain-based framework to make the citrus crop supply chain transparent and reliable in agriculture
(UMT Lahore, 2024) Mannan Ahmad Rasheed
Products such as Citrus vinegar, Citrus noodles, Citrus bread, and Citrus flour are made from Citrus, a vital crop for human nutrition and health. With more people worried about the safety of the food they eat, the citrus industry's supply chain has come into the spotlight. Furthermore, farmers can lose out on income from citrus crops even when they work tirelessly due to supply chain malpractices. Governments might potentially suffer huge financial losses as a result, especially during the off-seasons when citrus imports become costly. These problems stem from the Citrus supply chain's unreliability, lack of trust, transparency, and security. Our study introduces a Blockchain-based framework that can improve the Citrus crop supply chain in terms of security, integrity, reliability, and transparency. This will allow for full traceability of the crop from farm to fork. The Citrus Coin (CC) is a new cryptocurrency that will make it easier to record transactions between all parties involved across the Citrus supply chains. A crypto wallet, an economic model, and an ICO for the CC are all part of the suggested framework. We built a system for transaction processing that uses smart contracts to make citrus crop tracking and transparency easier, and to convert CC to fiat money. Additionally, to improve data security, openness and availability, we suggest using the Interplanetary File Systems (IPFS, which are) to store encrypted data of businesses, retailers, and farmers. Our suggested framework outperforms current supply chain solutions in terms of WI performance, as shown experimentally. This is especially true when looking at metrics like new block latency, average gas cost per transaction, and transaction verification time. Blockchain technology, supply chain management, smart contracts, citrus fruit, transparency and traceability, and food safety are all related terms.
A cloud based multi-agent system for traffic exigency services
(UMT Lahore, 2020) MUHAMMAD SALMAN ZAHEER
A multi-agent system can work on different tasks in a distributed environment to achieve single global goal. These systems can be integrated with Cloud computing to get better performance. A multi-agent system can perform more efficiently by utilizing cloud storage resources and computing power. Until now, a lot of research work has been done on traffic signal automation by using multi-agent systems. This research presents a new architecture by utilizing the power of multi-agent system and cloud computing to handle the exigency services in the urban transportation system. Exigency services are the special cases in the transportation system, which may include an emergency ambulance movement, police movement, Fire Brigade, and other exigent situations. In the proposed architecture, agents are deployed at each signal node, which collect information at each signal and transmit it to the cloud services. These agents are also able to handle the signal nodes in case of emergency. Agents at different nodes share information about exigency. Cloud service plays the backbone role with multi-agent while handling exigency. It analyses the input data and determines the presence of an emergency service. We have defined a priority mechanism to handle multiple emergency services at a particular crossroad or chowk. We also have presented different case studies to show the utility of the proposed architecture. It gives a better understanding of exigency service handling in an automated urban transportation system.
A conceptual multi-layer architrcture for nighttime pedestrian detection in autonomous vehicles by using deep reinforcement learning
(UMT Lahore, 2021) HARIS KHALID
The major challenge faced by autonomous vehicle today is driving through the busy road without getting into an accident especially with pedestrian. To prevent collision with pedestrian, vehicle requires the ability to communicate with pedestrian to understand their actions, motion change and intentions. The most challenging task in research on computer vision is to detect pedestrian actions, and intentions at night. The Advanced driverassistance systems (ADAS) has been developed for driving and parking support for vehicle to visualize, sense, send and receive information from the environment but it lacks to detect nighttime pedestrian actions and intentions. This article proposes a framework based on Deep Reinforcement Learning (DRL) by using Scale Invariant Faster Regionbased Convolutional Neural Networks (SIFRCNN) technologies to efficiently detect pedestrian gestures, postures and intention, through which the vehicle as agents train themselves from environment and forced to maximize the reward. The SIFRCNN has reduced the running time of detecting pedestrian action and intention from image by incorporating Region Proposal Network (RPN) computation. Furthermore, we have used Reinforcement learning (RL) in which the agent gets the state from the SIFRCNN and makes the optimize Q values and train itself to maximize the rewards. In addition, the latest incarnation of SIFRCNN achieves near real time object detection. Lastly, SIFRCNN has provided less miss rate and test rate and show effective performance in computational time and accuracy as compared to another available state of the art approaches in nighttime pedestrian actions and intentions detection.
A content base technique for the detection of sms spam data
(UMT.Lahore, 2018) Muhammad Farooq
Today text messages have a significant impact on our lives, but at the same time we face many critical problems due to SMS spamming. Therefore to detect spam messages and distinguish it via accurate filtering is a challenging task for researchers. In current research work content based spam SMS filtering technique is proposed depending on machine learning approach to distinguish spam messages firom mobile data while considering the low processing power and limited memory of mobile phones. From each text message; five attributes are extracted and based on these attributes/features, an unknown/unlabelled SMS message can be classified as spam or ham by a trained leaming algorithm. These attributes are the length of the text message and presence of repeatedly occurring spam words, count of spam words, combination of spam words and SMS class. It is shown that Decision Tree classifier performance is better than other machine leaming algorithms investigated. The other learning algorithms explored in this work are Naive Bayes and neural network architecture-Multilayer Perceptron Algorithm
a critical survey of s-box designs based on chaos theory
(UMT Lahore, 2021) HAIDER ALI BAIG
The data security, integrity and quality, in modern era of technology are very crucial for everyone whenever data transaction occurs. Various techniques are applied for securing the information and the optimistic way is by constructing of the Substitution Box (S-Box) that provides confusion in the data. Substitution process plays a vital role for constructing a strong cipher. An S-Box which shows high Nonlinearity, and low Linear and Differential Probability is marked as secure in the field of cryptography. For both user and the security related organizations, the main area of interest is on the phase of S-Box construction. As long as the S-Box shows strong properties like Linear Probability, Differential Probability, Non-linearity, Strict Avalanche Criterion and Bit Independent Criterion, it is nearly impossible for hacker to break into the information. By using substitution and permutation processes, which are the main phase of construction of S-Box. In this work, I have analyzed different types of substitution box techniques and provided a comparison between different techniques. Results of these techniques have been observed and a brief explanation is given that which S-box generating technique is best for the security of your message or information.
A detailed and comprehensive performance analysis on the largest covid-19 chest x-ray image dataset using a custom cnn model and state-of-the-art deep learning models
(UMT Lahore, 2022) ALI TARIQ NAGI
The modern Scientific World has ever been endeavoring towards battling and devising solutions for the newly arising pandemics. One such pandemic which has disarrayed the world’s accustomed routine upside down is COVID-19 and has devastated the world’s economy consuming around 45 million lives around the globe. The governments and scientists have taken front lines striving towards the diagnosis and engineering of the vaccination for the said virus. The COVID-19 can be diagnosed using Artificial Intelligence with an accuracy higher than the traditional methods using chest X-Rays. The subject research work involves an evaluation of the performance of Deep Learning models for COVID-19 diagnosis using chest X-Ray images on a dataset containing the largest number of COVID-19 images ever used in the literature research work according to the best of the authors’ knowledge. Further, a CNN model was developed, named Custom-Model in this study for an evaluation and comparison with the state-of-the-art deep learning models. The intention was not to develop a new highperforming deep learning model rather evaluate the performance of deep learning models on a larger COVID-19 chest X-Ray images’ dataset. Moreover, Xception and MobilNetV2 based models have also been used for evaluation purposes. The criteria for evaluation have been based on Accuracy, Precision, Recall, F1-Score, ROC curves, AUC, Confusion matrix, and Macro and Weighted averages. Among the deployed models, Xception turns out to be a top performer in terms of precision and accuracy while MobileNetV2 based model could detect slightly more COVID-19 cases than Xception and slightly fewer False Negatives while giving far more False Positives than the remaining models. Also, the custom CNN model exceeds MobileNetV2 model in terms of precision. The best Accuracy, Precision, Recall, and F1-Score out of these three models turn out to be 94.2%, 99%, 95%, and 97% as shown by the Xception model. Finally, It was found that the overall accuracy in the current evaluation was curtailed by approximately 2.2% as compared to the average accuracy of previous works on multi-class vii classification while a very high precision value has been observed which is of high scientific value.
A federated framework for air quality prediction in smart cities
(UMT Lahore, 2023) Shahan Uddin
Over the last couple of decades, due to the constant increase in urbanization and industrialization, the concern in terms of air pollution has become a serious concern. In most cities, the pollution in the air is mostly comprised of Nitrogen Dioxide (NO2), Ozone (O3), Carbon Monoxide and Particulate Matter; all of which can cause serious health issues. There is a very emergent need for a system that would detect the pollution in the air. This research presents a proposed framework that makes use of Federated Learning to not only lessen the communication overhead during the prediction process but also ensure the privacy of data. In order to ensure the effectiveness of the proposed work, the research will make use of Flower to simulate a federated smart city. The research will also make use of different Regression based Machine Learning algorithms such as Random Forest, Decision Tree, Support Vector Machine (SVM), and Logistic Regression for the training and evaluation of the research. Results from the simulations showed that Random Forest and Decision Tree performed better in comparison to a non-federated based simulation environment.
A federated learning based framework for citrus diseases detection
(UMT Lahore, 2024) Abdullah Mehboob
Increasing complexity in citrus diseases continues causing significant threats to global agriculture and bringing along several economic losses besides insecurity in food. The traditional centralized methods used for identification of diseases are seriously flawed due to problems of data privacy and computational inefficiency, and clear scalability issues across geographies. Whereas the deep learning models that are currently available are inherently so powerful, they require enormous datasets that cannot be centrally collected mainly for reasons of data privacy. That aims to solve such problems, a new federated learning framework has been proposed by drawing upon innovative hybrid CNN-ANN models which are highly accurate in the detection of citrus diseases. Using federated learning, which allows multiple clients to train a global model without sharing sensitive information, thereby maintaining privacy with improved performance. In the suggested model, our hybrid model is taking benefit from the extraction of spatial patterns by CNNs and sophisticated feature learning representation by ANN in classification of diseases. The proposed system attained a global test accuracy of 99.44% along with loss of 0.61, which further confirmed the reliability and strength of the proposed model be executed in real-world scenario also. Moreover, it's functionally efficient enough over distributed networks and scalable to large applications in the agricultural sector. Additional techniques of decentralized data processing and encryption strengthen the model's security. A new solution is provided for the detection of citrus disease through a privacy-preserving and scalable highly accurate real-time agricultural monitoring.
A framework for multiclass unusual activities recognition using deep learning
(UMT Lahore, 2023) Muhammad Ramzan
In the current era, automatic surveillance has become an important research challenge because of its numerous real-world implications for maintaining peace and order. Surveillance cameras have been used for monitoring human activities, behavior, and other changing information for over a decade. Recognizing human activities from surveillance videos has evolved into an active and progressive research area in computer vision and machine learning due to various challenges including environmental effects and various camera configurations. This study proposes a multiclass framework using deep learning for unusual activity recognition in video streams more efficiently and with better results than the existing system. The four basic types of activities are group-based activities, interactions, actions, and gestures. The activity's complexity and duration primarily determine this classification. The current research focuses on detecting unusual behaviors in three activities: actions, interactions, and group activities. In this research, the unusual behavior of the driver is chosen as a case study for the action category, unfair means in the exam for the interaction category, and violent activity detection for group activities. The state-of-the-art unusual activity detection techniques are explained in detail. Second, for each category, the existing methods are examined. Third, the pros and cons of various techniques are explored and comparatively analyzed. The outcome of these activities is discussed using various experimental setups and proposed models. The results of the first case study, driver drowsiness detection, are based on two experimental configurations. In the first experimentation, deep learning, machine learning, and hybrid model are proposed for unusual behavior recognition and classification. The YawDD publicly available dataset used, has an accuracy of 97.42 % for machine learning, 99.43 for deep learning, and 99.64 % for hybrid-based techniques. In the second experimental setup, various pre-trained models such as VGG19, ResNet101, VGG16, Inception-v3, and viii proposed CNN are utilized for real-time surveillance videos on newly created datasets. This dataset consists of five classes (Driver-Drinking, Driver-Eating, Driver-smoking, DriverNormal, and Driver-Calling). The accuracy achieved by ResNet101 89%, VGG-16 93 %, VGG-19 93.5%, Inception-V3 94% and proposed-CNN 95%. The second research problem, detect unfair means during exams, such as cheating activities among students has taken. The creation of the dataset is a considerable contribution to this research. The datasets include two types of exams: online and physical. The proposed Automatic Unusual Activity Recognition (AUAR) method uses motion-based frame extraction approaches to extract keyframes before applying advanced deep learning. We proposed 2D-CNN and 3D CNN architectures for physical exam activities and proposed CNN with pre-trained deep learning models for online exam activities to detect suspicious activities. For online exams, including four classes, accuracy was achieved: DenseNet 86 %, InceptionResnetV2 81%, Inception-V3 72%, and Proposed-CNN 94%. The YOLO-V5 has a few different performance measures, which include: precision 95.54%, mAP_0.5 95.40%, mAP_0.5:0.95 84.65%, and recall 93.16%. For the physical exam, Accuracy achieved 76% for machine learning based and 75 %, 77% for 2D-CNN and 3D-CNN models. In the group bases category, the proposed model includes pre-trained models and sequential CNN. The pre-trained models include Inception V4, ResNet-101, VGG16, and Inception_ResNet_V2 for classifying violent activities. The main assessments are performed using three standard available datasets: Hockey Fight, Movie’s dataset, and the Crowd Violence dataset (CVD). On the Hockey fight, Movie, and CVD datasets the system accuracy by using the proposed Sequential CNN obtained 98.26 %, 98.22 %, and 98.83 %. For Inception-V4, the results achieved 99.05%, 98.25%, and 97.70%, respectively. Finally, the thesis concludes with key considerations and future research directions
A framework for the evaluation & improvement of flowchart based programing environments
(UMT.Lahore, 2017) Nawara Latif
In the computer science field the core issue to learn programming, its difficult job especially for novice programmers. The process of learning programming is still slow and faced difficulties to the learners, despite having many advanced Integrated Development Environments (IDEs). Flowchart drawing environments introduced to minimize these difficulties to face the novices to learn programming. Many programming based flowchart environments are developed to teach programming with logics, algorithmic thinking and good opportunity for the learners to focus on acquiring problem skills without syntax error issues. Flowchart based programming environments used to teach programming an introductory course for novice learners, but the selection of best environment that is very important, according its major features, yet many efforts have been conducted to find the best environment for learners. Therefore, there is no proper method available to evaluate and compare these environments as to find the most suitable environment for learners. In this research, we have been proposed such a framework that evaluates these environments the suitable for novice learners at different stages, furthermore the framework have been involved a customized scoring function to compute the suitability score for these environments.
A framework for the evaluation & improvement of object oriented programming environments
(UMT.Lahore, 2017) Anam Ali
Learning a programming language for the first time is a daunting task, no matter what the age of the learner is. Novice learners or students, face many difficulties in learning programming at the beginning level. Many researches are made on this issue, but the object first approach considered the best method of learning programming. For teaching and learning Object oriented languages Programming environments are introduced which make programming easy and interesting. Choosing a programming environment for learning OOP languages at starting level is difficult. Many programming environments are object oriented like Alice, BlueJ, Scratch, Blockly and Greenfoot. These are used for both teaching and learning of programming languages. These OOP environments use physical objects and graphical structures to implement the programming principles and concepts. In this research, we have developed a framework compliance with OOD notations in UML and evaluated the features of OOP environments according to this framework. Furthermore, we evaluate the features of OOP environments and then devised a customizable scoring function and compute the suitability score of OOP environments.
A framework for the evaluation of safe city projects
(UMT.Lahore, 2019) Saba Javed Chattha
Safe City concept is comprised of providing a safe and protected environment for the residents of a city by the state authorities. Different models and concepts for this object are being used to make cities safer and crime free, all over the world. However, while studying and critically analyzing these Safe City Projects(SCP),it transpires that some SCP are up to the mark and delivering while some others are malfunctioning. Moreover, there is no proper evaluation mechanism framework to gauge the efficacy of these SCP. Since the unavailability of a standardized and encompassing mechanism, it is difficult to rank the best among the various prevalent SCP. In this research, an evaluation framework for Safe City Projects is designed to evaluate the different features of SCP according to this framework. Moreover, it will help in ranking the best SCP and will be able to improvise their working accordingly.
A framework to make wheat crop supply chain transparent and reliable using blockchain technology
(UMT Lahore, 2022) Zain Khalid Ansari
The wheat crop is the world’s second-largest cultivated crop. This crop fulfills almost 35% of human food demand and is used in making many by-products such as bread, flour, flakes, and other baked products. The World Health Organization (WHO) has been reported many issues related to food safety. The farmers have been putting a lot of effort into the production of wheat crop and their motivation is harmed by getting low revenues during crop harvesting season and the government suffers economic losses by spending additional money to import wheat from other countries in the off-season. These issues happened due to lack of transparency, security, reliability and traceability in the agriculture supply chain. Many systems have been developed for the agriculture supply chain safety, traceability, transparency and reliability however, all these systems have not been successful due to monopolistic centralized control. It has eventually gained consumer’s trust in branded products and rejects other products due to lack of traceable supply chain information. In this article, we have proposed a blockchain-based framework for supply chain traceability system that provides trustable, transparent, secure, and reliable services for the wheat crop. A crypto token called Wheat Coin (WC) has been introduced to keep track of transactions among the stakeholders of wheat supply chain. Moreover, we have introduced an Initial coin offering (ICO) of WC, crypto wallets and an economic model. Furthermore, a smart contract-based transaction system has been proposed for the transparency of wheat crop transactions and conversion of WC to fiat and vice versa. We have developed the Interplanetary File System (IPFS) to improve data availability, security, and transparency which stores encrypted private data of farmers, businesses, and merchants. Lastly, the results of the experiments show that the proposed framework improves as compared to previous crop supply chain solutions in terms of latency to add-blocks, per minute transactions, average gas charge for transaction, and transaction verification time
A gold standard business process model collection
(UMT.Lahore, 2015) Rao Faizan Ali
Business Process modeling enables an organization to analyze their business operations and the change comes into those operations with the passage of time. There are too many processes in an organization; for managing those processes organizations maintain repositories of process models. However process models repositories are build in different stages and by different people in the large organizations so with that case repository must have data duplication and noisy data. For managing that data there must be some automated technique required.To managing those noisy collections researchers are trying to build repositories which are advanced and manageable easily. For making those kind of repositories there are some advanced features required .these features are classified into four types, evaluation, comparison, management and presentation. Whereas evaluation covers the analysis of quality, correctness and performance analysis .comparison considers similarity between models and pattern based analysis. Management describes about how to control extensions, individualization and reference based completion of the repository. Last and most useful feature is presentation of the process models presentation relates to useful generalization of processes and secondary notations (size, color,etc).As people have tried to invent several automatic techniques for matching, it is very important to rigorously evaluate these approaches. If the approaches are not evaluated rigorously, there are chances that the approaches our work for process models of specific category, such as large process models or dense process models. Also, one would like to be able to say which are better, which worse, and also which words, or varieties of issues, present particular problems to which algorithms. In order to evaluate approaches to process matching/detection it is useful to have access to a collection of models that containing examples of the types of models/models of different features that we aim to identify. To assist in this study it was necessary to create a comparable corpus consisting of a selection of process models that have specific features.The purpose of this research was to develop a gold standard collection of process models, classification of process models, develop queries on defined classification and matching those queries to determine validity of collection. In our research work, we have built a collection of process models which are in BPMN 2.0. After creating the corpus of process models we classified them. The task of classification will be done based on structural Metrics like size, type etc. A set of queries will be made from each class then we performed query matching on the collection of process models. For process models matching Manual approaches of similarity computation used.
A moel for the prediction of ozone layer weakness using fuzzy logic
(UMT Lahore, 2022) SIDRA IKRAM
Since the evolution of mankind, man has always been curious about why and how things happen. Why is something the way it is, the question still arises. More, humans have been trying to invent and discover new things one better than the other in every domain. With time this human behavior had been increasing as more and more is being invented and discovered day by day but the quest is continued. The proposed method is developed in a way that it uses the ozone layer weakness prediction data from a specific population related to the factors that are economy, pollution, and social setup as input and then it analyzes the data and gives an output in the form of OLWPSF (Ozone layer weakness Prediction System Using Fuzzy Logic). This system is developed to find out the expected burden on the ozone layer due to the above-mentioned factors by finding out their effect on the weakness of the layer of a certain region or country. The output results thus will be available for the use of government officials, policymakers, and authorized personnel in that region so that they would be able to make decisions based on those results.
A neural machine translation pipeline for translation of english based sentences to pakistan sign language sentences
(UMT Lahore, 2023) Aamina Khan
With the increasing need for communication in universities, organizations and daily life, numerous difficulties are faced by the hearing impaired. Neural Machine Translation (NMT) is a computational linguistics application of deep learning that performs machine translation using artificial neural networks. It predicts the sequence of words as they are translated from one language to another. For a model capable of such translation, extensive datasets containing translated sentences in both languages are used for training purposes. The architecture is built upon encoder decoder structures that perform the actual translation. This translation algorithm is widely used by industries along with the recent development of attention mechanism that enhances the accuracy of the models. In this work, I have developed Pakistan Sign Language (PSL) dataset of approximately 87K sentences and trained it on two different Neural Network Models: Recurrent Neural Network (RNN) and Transformer. The RNN model is given one word at a time and generates one word at each time step. To compare the performances of both the models using distinct word embedding we utilized The Bilingual Evaluation Understudy (BLEU) Score, Word Error Rate (WER). The BLEU score automates the human evaluation process which accelerates the PSL generation. On a scale of 0 to 1, 1 being the closest to human reference translation, it identifies the dissimilarity among human translation and machine translation. When compared it was concluded that Transformer shows slightly better results. The achieve accuracy of Transformer is closer to that of human predictions.
A novel approach for the classification of alzheimer disease using machine learning techniques
(UMT Lahore, 2021) KAMRAN ARSHAD
Alzheimer's disease (AD) is an irreversible, complex mental disorder that affects individual thinking, memory, and skills. Many different statistical methods and algorithms are often employed in the study of neuroimaging data to classify Alzheimer patients (AD) and normal control subjects (NC). This study aims to classify different levels of Alzheimer patients and Normal control subjects by using an ML model. Machine learning (ML) has allowed doctors to achieve remarkable results, and healthcare is using ML-based models to detect stage or level in patients. This allows analyzing the healthcare data and uses the traditional computer-aided detection (CAD) to assess breast cancer. Machine learning has become an accepted clinical practice and allows doctors to evaluate the ML model to detect AD at an early stage. A major aim is to diagnose patients having AD by analyzing the data of patients and classifying them into four stages, having diagnosis results as 0 = Normal, 1 = Early-AD, 2 = Mild- AD, 3 = Sever-AD based upon his/her features i.e. its Occipital lobe, Frontal lobe, Parietal lobe, Amygdala and Hippocampus. In this research different machine learning algorithms random forest, logistic regression, KNN, SVM, decision tree, and MLP are used to classify patients as Alzheimer patients (AD) or normal control subjects (NC). The private data set is used for applying these algorithms to get the best accuracy. So, for all the above-mentioned algorithms Random forest (RF) and KNN is one of the more efficient and accurate algorithms to classify the stages of AD. And here also fitted the matthews_corrcoef for Random forest (RF) and KNN is 0.99% and accuracy score is 0.99%.

Browse

Browsing MS by Title

Results Per Page

Sort Options