Browse
Recent Submissions
Item A MINI-FRAMEWORK FOR AN EFFECTIVE SCRUM DEVELOPMENT ENVIROMENT(UMT, Lahore, 2018) MARIYAM IFTIKHARSoftware process models are being used by software industries and agile is considered the best methodology for better software development. Scrum framework is the specialized version of agile model. Software industries considered scrum as most adopted framework for successful software development. Currently scrum is adopted by both local and international level software houses. The reason behind the increasing demand of this framework is its fast response to user needs, better productivity, upgraded cooperation and quick delivery of products. Numerous challenges are faced by the practitioners while implementing scrum. The experts have contributed in removing flaws, issues and limitations to make this framework attractive, so that it can be adapted by software industries, however there is no solo effort that reports all such issues. As the scrum team are facing some issues in relation to the development of product, a framework is required to increase the effectiveness of scrum at local level software houses. This research study intends to present a specialized version of scrum so that all issues and drawbacks faced by the scrum team can be encountered and team efforts can be improvised, leading to better performance and visibility of work pertaining to software development. The proposed framework is increasing the effectiveness of traditional scrum framework without disturbing the execution time, integrity and simplicity of this framework. This framework can simplify the team efforts and ease the processes for analyzing the productivity of desired quality product in scrum.Item DENDROCHRONOLOGY: ESTIMATING TREE LIFE BASED UPON TREE RINGS(UMT, Lahore, 2018) SHADAB HASHMATNatural disasters affect the economy and society of a country, furthermore they are also caused by the weather. The understanding of past climate conditions help us predict natural disasters over a period of years. Scientists and researchers are doing lots of work in the domain of predicting climate condition. From all domains, information that extracts from natural evidences like ice core, tree-ring and verve (rock) give more precise result while predicting climate and natural disasters. Tree rings pattern are biggest source of information than others in term of predicting climate changes. The method for analyzing pattern of tree-rings or study tree-rings is known as dendrochronology. Dendrochronology is scientific method for dating tree-rings. There are various methods and techniques proposed for dendrochronology but still the methods have to be improved for better performance. In this thesis, we create an automatic computerized system for dendrochronology based on computer vision techniques. This system estimates tree’s age in three phases including preprocessing, tree-ring detection and determining tree’s age. In preprocessing phase, we convert image to grayscale for size reduction and for noise removal and smoothness, Gaussian filtering is used. For tree-ring detection, we used Sobel edge detector and for estimating age proposed a novel technique which determine trees age by counting number of continuous consecutive black and white pixels in edge image. For this research, we created own dataset of 120 images of horizontal cross section of tree-rings from different sites. We conducted experiment on that dataset and obtained 90.84% accuracy of tree-ring detection. The other main contribution of this thesis is generating labeled dataset which facilitate future researches.Item CUSTOMISING SCRUM PROCESS TO IMPROVE GLOBAL SOFTWARE DOCUMENTATION(UMT, Lahore, 2018) ANILA LIAQATItem Career and Skills Recommendations Using Data Mining Technique: Matching Right People for Right Profession, in Pakistani Context(UMT, Lahore, 2018) Hafiza Maria KiranNowadays, recommendation systems are commonly used by the people for finding the products which best match with their individual preferences. In the context of profession and recommendations, a lot of recommendations systems are available on the internet for the help of jobseekers. The systems are performing well and generating job recommendations for people jobs but they have some serious problems that are faced by jobseekers in Pakistan. They are not much intelligent, require a lot of user’s time in filling long forms for registration so that they can get recommendations. Moreover they do not give suggestions that which skills will be suitable for a specific profession. A user must has to spend too much time for applying a job and still he doesn’t know the skill that is more valuable for him and is it the best job for his skills on which he is going to apply? The problem is that people are not clear in which field they should start or switch working. Actually there is a point that first of all one should be clear about his/her profession and important skills regarding selected profession in which he/she wants to start a career, then he/she should start finding job related to the selected profession. Based on above issues, there is a need to design such a system that can overcome the problem of profession selection and skills suggestions so that it can be easy for a jobseeker to apply for a specific job. In this research, the problem which is discussed above regarding profession and skills recommendations is resolved by proposing a model by using Association Rules Mining, a data mining technique. In this model, professions are recommended to job seekers by matching the profile of applicant or job seeker with those persons who have same profile like educational background, professional skills and the type of jobs which they are doing. The data collected for this research itself is a major contribution as we collected it from different sources. We will make this data publically available for others so that they can use for further research.Item AN ADAPTIVE FRAMEWORK FOR ROBUST ONLINE VOTING SYSTEM(UMT, Lahore, 2018) AMBER NAWAZA standard and protected online system is required to conduct elections in contemporary era to make voters assured. Currently, each country is using its own electoral process. Inconsistency of poll results has created various problems. Numerous models have been adopted to address the problems of security, privacy, validation and quality control. Yet, these initials do not encircle all of the basic needs for a full organization. Exploring Online voting from a systems approach can show the commonalities of the current arrangements and the possible solutions in the balloting procedure. It has always been a gigantic task for the Election Commission of Pakistan to conduct free and fair elections in our state. Billions of rupees have been spent on this phenomena to make sure that the elections are corruption free. However, nowadays it has become common for some entities to incorporate in the rigging which may eventually lead to a result opposite to the actual will of the people. This thesis aims to provide a new voting system using electronic system in order to avoid manipulation and to maximize the accuracy of the system.Item Human Fall Detection(UMT, Lahore, 2018) Reamsha KhanFall-induced damages are conjoint in the old populace. Postponement or short-age of medical precaution after the incident of a collapse often causes damages, in some cases it became so intense that it may result in death of the victim. Hence falls are serious incidences for the aged person. For this problem automatic detection of fall on the spot can play a vital role in timely medication care which ultimately helps to de-crease the medical complexity. Keeping in view the above stated crucial problem, in this paper we will de ne an e cient and e ective system which can detect the fall centered on dataset of videos produced by means of numerous cameras.This research proposed an approach which perform better in terms of accuracy as related to the additional present approaches.It utilize numerous descriptors of image or various features which are sus-tained to various training classi ers to recognize human falls.Item SEQUENCE-BASED PREDICTION OF MULTIPLE LIPID MODIFICATION SITES IN PROTEINS BY INTEGRATION OF PSEAAC AND STATISTICAL MOMENTS(UMT, Lahore, 2018) WAQAR HUSSAINLipid modification of a protein, which can be co-translational or post-translational, is known for regulation of various physiological factors, such as protein-membrane interactions, protein-protein interactions, protein stabilization and enzymatic functionality. Due to the association of these lipid modification sites with various diseases, its timely prediction can help in diagnosing and controlling the associated fatal diseases. Here, we present a method for prediction of multiple lipid modification sites, in which we have incorporated PseAAC with statistical moments for the prediction. The aim of this study is to propose a new and more accurate predictor for lipid modification sites, based on the 5-step rule, to make it easier for the experimental scientists getting desired results. A benchmark dataset of 893 positive and 1093 negative samples for NMyristoylG-PseAAC, 90 positive and 100 negative samples for SFarnesylC-PseAAC, 74 positive and 100 negative samples for SGeranylgeranylC¬PseAAC, and 436 positive and 500 negative samples for SPalmitoylC -PseAAC, is collected and used in this study. For feature vector, various position and composition relative features along with the statistical moments are calculated. Later on, a back propagation neural network is trained using feature vectors and scaled conjugate gradient descent with adaptive learning is used as an optimizer. Self-consistency testing and 10-fold cross-validation are performed to evaluate the performance of predictors, using accuracy metrics. For self-consistency testing of NMyristoylG-PseAAC, 96.93% Acc, 97.09% Sp, 96.80% Sn and 0.94 MCC is observed, whereas, for 10-fold cross validation 94.41% Acc, 94.06% Sp, 94.70% Sn and 0.89 MCC is observed. For self-consistency testing of SFarnesylC-PseAAC, 95.79% Acc, 96.67% Sp, 95.00% Sn and 0.92 MCC is observed, whereas, for 10-fold cross validation 93.68% Acc, 95.56% Sp, 92.00% Sn and 0.87 MCC is observed. For self-consistency testing of SGeranylgeranylC-PseAAC, 95.91% Acc, 95.77% Sp, 96.00% Sn and 0.92 MCC is observed, whereas, for 10-fold cross validation 92.98% Acc, 92.96% Sp, 93.00% Sn and 0.86 MCC is observed. For self-consistency testing of SPalmitoylC-PseAAC, 98.08% Acc, 98.62% Sp, 97.60% Sn and 0.96 MCC is observed, whereas, for 10-fold cross validation 94.66% Acc, 96.79% Sp, 92.80% Sn and 0.89 MCC is observed. Thus the proposed predictor can help in predicting the targeted lipid modification sites in an efficient and accurate way.Item SKETCH RECOGNITION USING DEEP CONVOLUTIONAL NEURAL NETWORK(UMT, Lahore, 2018) MUHAMMAD ZAKI SIDDIQUISketch Recognition and Classification has received a great deal of attention in the last few decades. Mankind is using sketches for communications from ancient times. Even in today’s world, when people of two different languages meet, they use gestures and sketches occasionally. These sketches draws by non-professionals. As a result, great level of abstraction which makes it more challenging problem than the image recognition in machine learning. Also, sketches contain fewer features like colors or backgrounds. Due which two totally different objects could have the resemblance. This increases the difficulty for recognition as well. Different terminologies have been used over the years to achieve this objective. Major methodologies includes, image processing, computer vision, machine learning and most recently deep learning. Deep learning is the more focused type of machine learning which use deep architecture based on several layers which eventually helps it to perform better in recognition and classification tasks. The convolutional neural network (CNN) has proven as a worth full approach in vast range of applications for identification and classification task in the visual domain (images). This paper proposes a technique for sketch recognition based on CNN. We utilize a publically available database from Eitz et al (Mathias Eitz, James Hays, Marc Alexa. 2012), which work as a benchmark for object sketch recognition which compromises of 250 classes and total 20,000 images. Most of the previously conducted experiments on TU – Berlin dataset was based on the pre trained models. And they attempt to improve the results by changing the few parameters. We present a minimal model for the identification of sketches. We build the model from very scratch to eliminate any additional complications in it. Due to the hardware limitation, we conduct our experiment on one – fifth of the dataset. And able to achieve 76.96 % accuracy. We analyze our model performance under different hyper-parameters and discussed the challenges that we faced during the experiment.Item DESCRIPTIVE ANALYSIS OF PAKISTANI CRIME DATA USING DATA MINING TECHNIQUES(UMT, Lahore, 2018) FARIA FEROOZCrime rate has increased in many cities of Pakistan. In order to reduce crime rate, there is a need to understand and analyze emerging patterns of criminal activities. This study analyzes crime rate by using the crime dataset of Lahore City, Pakistan. The analysis on dataset is made through data visualization, identification of similar characteristics through Clustering and frequent patterns are extracted by using Association rule mining. Hierarchy of crimes is developed, top three crimes and locations are identified by data visualization. K-means clustering algorithm is used to group similar characteristics of crime related events and find out possible risk factors on other locations. Apriori algorithm of Association Rule Mining is used for extraction of frequent rules or patterns from data. Numbers of frequent crime pattern rules are extracted and analyzed. The proposed solution can help police department to reduce crime incidents and detect crime event situations on early stages.Item THE ROLE OF SOCIAL MEDIA AND BIG DATA IN POLITICS(UMT, Lahore, 2018) QURATULAINThe use of social media has increased tremendously in the last few years and this use is producing trillion of amounts of data every day. This large amount of data from social media platforms enabled researchers to study the patterns of data and extract useful information from that data. In 2008, the social media data was used for election prediction and to persuade the citizens which were a huge success but the vital information was processed and social media analysis was being performed manually. Different countries including Pakistan took the same initiative as well and run very successful campaigns. Some researchers claimed that it is possible to predict and influence the election results using different social media platforms and used different techniques to prove their claims. The techniques fall un- der ten different approaches: surveys, machine learning, statistical, data mining, dictionary based, linguistic anthropology, ontology, software based, hand-coded and case-study based. Among all these approaches, machine learning approaches outperformed and hence gained a lot of attention from the research community. To analyze the role of social media in politics, this paper presents the sentiment analysis that was used to classify and predict election results. The test results demonstrate that the social media data can be utilized for inferring political behaviors of various parties. Positive, negative and neutral conduct of the followers of a certain party and party’s campaign effect can be anticipated from the analysis. The analytical results demonstrate impressive correspondence with actual results as published by Election Commission of Pakistan.Item STUDENTS’ PERFORMANCE PREDICTION BASED ON COURSE DIFFICULTY, TEACHERS’ GRADING BEHAVIOUR AND STUDENTS’ PERVIOUS PERFORMANCE(UMT, Lahore, 2018) MUHAMMAD TAYYAB MIRImportance of education is not only necessary for students but it also plays vital role for bright nations. Good students have high chances to have bright future. Students’ performance in their professional degree is a reliable tool which reflect their abilities in their domain. These performances need to be observed carefully in order to identify if a student’s performance is dropping sequentially, suddenly or is below average continuously throughout the degree. If it happens, than such students need to be guided by advisors and teachers and extra efforts should be applied to improve the performance. The aim of this study is to predict a student’s performance before it actually drops down at the end of the semester. Prediction of students’ performance at the start of the semester will help to take necessary precautions to avoid poor results and dropouts. In our study, we have proposed multiple supervised learning techniques of data mining for predicting student’s performance using regression analysis and classification on unlabeled and labeled data set respectively. Researchers have already predicted student grades in courses and predict their performances using previous performances and other factors but teacher’s grading behavior and courses difficulty are two major factors in student’s performance. These factors were never considered in predicting student’s performance before. The proposed supervised learning techniques show remarkable result with up to 94% of accuracy. These results are verified by comparing with actual GPA of the students.Item LEARNING BASED APPROACH FOR DISTRACTION DETECTION OF A DRIVER USING HEAD POSE ESTIMATION SPECIFICALLY HEAD PANNING FOR SAFETY DRIVING(UMT, Lahore, 2018) Halima JamilThe number of fatalities caused due to traffic accidents is very high. The biggest cause of distracted driving fatalities is a driver’s mind meandering for long enough leading to an impact. Therefore, researchers have been working on automated systems for monitoring the driver intentions. This survey provides a chronological evolution of a driver distraction detection techniques. The contribution of this survey is three fold. Firstly, it summarized different algorithms and techniques. Secondly, it deals the research conducted into different eras. Thirdly, it provides the datasets available so far.Item Prediction and Classification of Ransomware and Goodware based on Static Analysis using Neural Network Approaches(UMT, Lahore, 2018) SAAMAN NADEEMWith the increase in development of technology, the most common threat to our society is “Hacking”. With lack of security, there are greater number of chances for hackers to attack the system or to hijack it. Nowadays hackers are utilizing many different and complicated techniques, which have created a difficult challenge for the victim to distinguish the malicious content with traditional malware recognition methods. Ransomware is the latest trend of blackmailing the system’s users. Ransomware encrypts the user’s data or locks the system and demands payment, in order to decrypt it. Many machine learning approaches have been proposed for detection of ransomware, but these lack in detecting ransomware on time, which results in data inaccessibility. This research proposes two approaches based on static analysis of ransomware and, is specifically designed to predict and classify ransomware on the user’s system. It starts by predicting the ransomware from the available data set. Further, designed framework using machine learning approaches for prediction and classification of ransomware, two experiments have been implemented. The machine learning techniques include Artificial Neural Network (ANN) and Deep Neural Network (DNN). The proposed framework monitors a set of activities performed by ransomware. A dataset containing 582 ransomware samples, which belong to 11 different families. Also, 982 samples belong to Good ware. The proposed model is trained and tested on the given dataset. The proposed classification ANN model shows the accuracy of 98.56% for detecting and classifying ransomware and DNN shows the accuracy of 99.06%. Hence, these results suggest that by using Neural Network based approaches and using the support of static analysis, ransomware prediction can be done. As ransomware samples show that at running time ransomware holds the same features belonging to different families. The proposed framework will also help in predicting new ransomware having similar features.Item CAPSULE NEURAL NETWORK FOR AUTOMATIC DETECTION OF DIABETIC RETINOPATHY(UMT, Lahore, 2018) RUHAMA SARDARDiabetic Retinopathy (DR) is a disease which becomes the cause of blindness and visual impairment usually in middle-aged patients. It is estimated that over 93 million people are affected by DR. Detection of Diabetic Retinopathy manually is a very time consuming and expensive task, which requires trained ophthalmologists to examine and evaluate DR using digital fundus photographs of the retina. Many researchers are building CAD techniques from many decades for timely detection and reducing the burden of ophthalmologists. Deep learning techniques have boosted the performance of fundus diabetic retinopathy (DR) image classification. More precisely, convolutional neural network (CNN) achieves superior performance to that of the conventional machine or deep learning algorithms. Recently, in November 2017, a novel type of neural network named capsule network (CapsNet) was introduced to overcome the shortcoming of traditional CNN models. In this thesis research, we present three CapsNet architectures with limited samples for five stage DR classification, which is inspired by the simplicity and comparability of the shallower deep learning models. The presented CapsNet architectures were trained using publicly available Kaggle dataset. The experimental results show that CapsNet shows convergence behavior and better accuracy for the complex dataset. For CapsNet by using the kaggle dataset, achieves 76% accuracy, 69% sensitivity, 83% specificity respectively. This can be considered as better performance for such a newly developed model which lack much important information till now. Moreover, we observed that training the CapsNet model requires significant computational resources and its performance falls below the average performance level of CNN. However, we argue that CapsNet seems to be a promising technique for overcoming the limitation of CNN. Further by using more robust computational resources and refine CapsNet architectures, the better performance can be achieved but still, our proposed system can be used to diagnose DR in early stages and assist in the grading of diabetic retinopathy.Item IMPLEMENTING FINAL YEAR PROJECTS USING SCRUM APPROACH - A GUIDELINE(UMT, Lahore, 2018) Agha Azeem Ur Rehman KhanItem Ad hoc Collaboration Space for Distributed Cross Device Mobile Application Development Using WiFi Direct (ACS)(UMT, Lahore, 2018) Imran Abbas KhawajaOver the last few years, we have seen an enormous increase in the usage of electronic devices smart phones, tablets, laptops, TVs and wearables, which are developed by different manufacturer for different platforms. People surrounded by these devices need to interact with them during the meeting, presentation, class room and lots of other collaborative activities to share and receive information across the devices. However, the interaction among these devices is still device centric and dependent on the expensive fixed software and hardware infrastructure. In a situation, where fixed infrastructure service does not exist, suspended or disrupted due to some reasons, the interaction across these devices is not possible. In this study we have presented the framework that provides the highly performing reliable ad hoc network as well as facilitate the development of cross device distributed mobile applications without using any fixed expensive infrastructure. The novelty of our approach is that it has reduced the application development time by hiding the complexities and providing the easy to use API’s for application developer to build distributed mobile applications. Otherwise, the development takes extra time and distract the developer to build cross device applications due to high complexity level. This framework supports diverse types of android based devices of different manufacturer and all interactions are done using the WiFi Direct that enables the nearby devices to communicate with each other without the need of any common access pointItem A SCALE AND ROTA OIT N I VN ARIANT LC AS EIFIS R OF R ITPO CA C L HARA TC ER REC GO N ITI ON RU FO UD NASTA QIL UE OF NT(UMT, Lahore, 2018) Khawaja Ubaid UR RehmanItem LEARNING BASED APPROACH FOR RICE GRAIN CLASSIFICATION USING SHAPE CUES AND MORPHOLOGICAL PARAMETERS(UMT, Lahore, 2018) SHAMAILA ISLAMRice is a very important staple food fulfilling the need of large number of people around the globe. There exist different types of rice with many qualities to match different consumer preferences. Rice grain quality factor like grain width, length, height, texture etc. may also vary among different types of rice super kernel, basmati, and kernel etc. In whole world, there exist 40,000 varieties of rice. This paper use total sixty pe of rice grain. we aim to do classification of rice using different datasets by applying feature extraction based approach using various classifier including Support Vector Machine (SVM) that give 95.122 percent accuracy and Naive Bayes give 91.845 percent accuracy and Neural Network (NN) give 97.541 percent accuracy and decision table(DT)give100percentaccuracyandRandomforest(RF)give 95.318 accuracy and J48 give 100 percent accuracy that give best results. After done with training of datasets, derive results by testing it. The main aim of this thesis is to compute various morphological parameters including length, width, area and curvlet of rice grain that includes point of boundaries of rice that give the location than find out the direction of points on rice do take slope and perform their classification. we do pattern recognition for different type of rice grain draw BOX-PLOT that tells the types according to length of rice.Item POSE INVARIANT FACE RECOGNITION SYSTEM(UMT, Lahore, 2018) SHEIKH BILAL AHMEDComputer vision systems open a new challenge to recognize human faces under varied poses in similar capacity and capability as human-beings perform naturally. For surveillance applications, Pose-Invariant Face Recognition (PIFR) will become a major break-through by presenting the solution of this unique challenge. In recent decades, several techniques are presented to address this challenge over well-known data-sets. These efforts are divided chronologically into seven different approaches say geometric, statistical, holistic, template, supervised learning, unsupervised learning and deep learning. Among these deep learning techniques have shown more promising results and have gained attention for future research. By reviewing PIFR, it is historically divided into five eras based on 160 referred papers and their cumulative citations. This paper proposes a new PIFR method which uses metrics of the facial features to identify an individual’s face under any arbitrary pose. With the help of BoRMaN, the face region is detected and facial points are pinpointed. Once the facial points are located, the important facial features are acquired and a geometric feature set is created which contains various metrics of the features. These feature sets are used to train and test boosted SVM classifier. Experiments are conducted on three different databases, FERET, CMU-PIE and TEXAS. Results show that the proposed method delivered an outstanding performance on three databases by delivering unexceptional accuracy rates over existing works. Furthermore, the conclusion and future works are also presented.Item ON THE FRONTIERS OF BASIC AND COMPOUND EMOTION(UMT, Lahore, 2018) Saira NazAutomatic facial recognition is a challenging task due to the various categories of emotions. It has gained a lot of attention by researchers in multimedia retrieval and human computer interaction and related disciplines. The human face is the natural way of non-verbal communication for human interaction. In real life scenario, facial expressions and emotions are nothing but responses to the external and internal events of human being. Although, for the last decades, there has been immense development and mining of static images and dynamic images techniques for basic emotion and compound emotions detection but still, more work needs to be done. Every technique has its own uniqueness, performance and limitations. A lot of researches have been done regarding basic emotion and compound emotion detection that needs to be reviewed and summarized. This thesis is consisted of four main parts. The first is the encapsulation of the research papers of basic and compound emotion detection according to different algorithms & techniques. Second is the distribution of research papers of basic and compound emotion detection in the form of different eras. Third it discusses the data-sets that are used in different research papers. Fourth is the proposed methodology to detect and recognized the basic and compound emotions with the help of geometric features. Boosted classifiers are used for classification