Research Projects
Potentials of Cooperative Clustering
February 2014 – May 2014Cooperative Clustering on Graphs (CC/G) is a new graph-based clustering approach that can be useful for addressing many graph-mining research challenges. It can identify anomalies in graph-based datasets, which is applicable in fields including but not limited to network intrusion, fraud detection, health monitoring, and sensor networks. It can also be used as a technique for subgraph matching in the context of dynamic graphs and datasets that represent the temporal evolution of relationships between entities.
Cooperative Based Software Clustering on Dependency Graphs
June 2012 – June 2014This research project came with a novel ensemble clustering approach that synthesizes a new solution from the outcomes of multiple constituent clustering algorithms.
Growth Modeling and Prediction of Software Reliability
June 2013The goal of this project was to support top management who are deciding whether software or a specific product is ready for release. A release date is predicted based on failure data obtained during the testing completed and on a given target reliability measure.
Human Gesture Recognition (Research Project)
September 2010 – June 2012This research considers the task of labelling videos containing human motion with action classes and distinguishes between normal and abnormal actions.
Use of HMM and Affine Optical Flow in Online Video Categorization
April 2011Today’s content-based video retrieval technologies are still far from satisfactory, fundamentally because of a lack of content representation that bridges the gap between the visual features and the semantic conception of the videos, especially in real time. This work involved the development of a proposed motion pattern descriptor based on affine optical flow that characterizes motion generically. This representation then led to the design of a model-based classification scheme that effectively maps video clips to semantic categories. A hidden Markov Model (HMM) is used as the classifier. Experimental evaluation revealed that the new scheme also significantly improves the performance of motion-based retrieval due to the comprehensiveness and effectiveness of its motion pattern descriptor and its semantic classification capability.
Author Identification using Compression and Machine Learning
April 2011This project entailed a comparison of two paradigms for author identification: the first is based on compression algorithms that avoid the entire process of defining and extracting features to train a classifier, and the second takes into account a classical pattern recognition framework that relies on proposed linguistic features to train both a support vector machine (SVM) and naive Bayes classifiers. Comprehensive experiments conducted with a dataset composed of 5 writers showed that compression algorithms may provide superior performance. The report of this project includes a discussion of the advantages and drawbacks of both paradigms.
Construction Cost Estimation using Case Based Reasoning
2004Using VB and MS-Access, I designed and implemented a cost-estimation software tool using case-based reasoning. This tool was used for calculating shop rates based on input comprised of direct labor, machine price, number of employees, and gross sales, combined with 11default factors, including overhead costs.
Construction Cost Estimation using Case Based Reasoning
2004Using VB and MS-Access, I designed and implemented a cost-estimation software tool using case-based reasoning. This tool was used for calculating shop rates based on input comprised of direct labor, machine price, number of employees, and gross sales, combined with 11default factors, including overhead costs.
