Hello world! I am Shazan Jabbar, a Data Scientist at Alberta Machine Intelligence Institute (amii). I am also a part of the DoMiNO research team. I have a Master's in Computing Science from the Department of Computing Science at University of Alberta under the supervision of Prof. Osmar Zaiane. I also closely work with Prof. Alvaro Vargas.

I enjoy doing research in interdiscplinary projects and applying data mining methods to solve real world problems. During my Master's study I worked with DoMiNO team, developing spatial pattern mining techniques to discover interesting associations between air pollution and adverse health effects in Canada. I also developed data visualization tools to help our interdiscplinary collaborators and stakeholders at DoMiNO team to explore the patterns I discovered. My Mater's thesis is mainly based on these work. Currently I continue my work as a full time researcher at amii and DoMiNO, extending some of my methods, building predictive models, writing papers and developing the data visualization tool VizAR.

I also love coding. During my Master's I also worked as a professional programmer for amii in an Information Retrieval project. Inbetween my Master's and Bachelor's I worked as a Software Engineer (Databases and ETL) for a Syllicon Valley based startup called Leapset (Pvt.) Ltd. (now known as Cake Corporation).

My current research interests are in association rule mining, graph/network analytics, deep learning, topic models, supervised learning models such as classification & regression analysis, information retrieval & text analytics, and spatial data mining.

My Erdos number is 3! (I co-authored with Dr. Russel Greiner. Dr. Greiner co-authored with Dr. Michael Molloy and Dr. Molloy co-authored with Dr. Paul Erdos). You can find more about me in my resume, LinkedIn or in Scholar! You can also contact me via my email: mohomedj[at]ualberta[dot]ca.

Education

University of Alberta, Edmonton, Canada

Master of Science in Computing Science, GPA 4.0/4.0

Thesis: Discovering Spatial Patterns using Statistically Significant Dependencies [Link]
Advisor: Prof. Osmar R. Zaiane

University of Colombo, Colombo, Sri Lanka

Bachelor of Science in Computer Science, First Class Honors

Thesis: PageRank Based Core-Attachment Model to Detect Protein Complexes by Analysing Protein Networks [Link]
Advisor: Dr. Ruvan Weerasinghe

Experience

Data Scientist

Alberta Machine Intelligence Institute, Canada

Spatial Data Mining, Data Visualization, Machine Learning, Statistical Pattern Recognition, Information Retrieval, NLP

Graduate Research / Teaching Assistant

University of Alberta, Canada

Data Mining, Machine Learning

Software Engineer (Database / ETL)

Sysco Labs (A Sysco Company)

Database Management, Data Integration, Report Generation, ETL Job Designing

Research / Projects

DoMiNO (Data Mining for Neonatal Outcome) [MSc Thesis]

This project aims to discover association patterns between air pollutant emissions and adverse health conditions in Canada. To address this goal we followed a spatial data mining approach and proposed a new technique to transform a spatial dataset into a transaction dataset to apply association analysis techniques more easily. We proposed two novel pattern discovery techniques to find spatial associations using statistical significance tests. One such proposed technique discovers a novel type of spatial patterns called spatial contrast sets, which aims to characterize a particular spatial group and contrast it from the others. The other type of pattern, spatial common sets aims to find association patterns which are common among many spatial groups. These patterns were able to discover many interesting associations between air pollutants and adverse health conditions in Canada successfully. Currently we are implementing a pattern visualization software suite, VizAR, to help practitioners and our collaborators to explore the patterns we discover.

Advisor: Prof. Osmar R. Zaiane

Golden Retriever (Question Retrieval System)

Proposed a community question retrieval system based on a weighted TF-IDF scheme, relevance heuristics and term expansions. Goal of this system is, when given a search query, to find similar questions in community forums. Compared to other well known methods such as LDA, LSI, Language Model based IR techniques, BM25 and Word2Vec based models our approach performs well in retrieving similar questions in community forums. This proposed model was placed among the finalists of IEEE International Conference on Helath Informatics Data Analytics Challenge in 2015

Advisor: Dr. Osmar R. Zaiane

OPECH - (Extracting Protein Clusters from Hierarchies)

Proposed an algorithm OPECH which builds a cluster hierarchy of proteins based on their topological distance in protein interaction networks and uses a semi-supervised schoring scheme in combination with FOSC framework to extract optimal set of clusters without defining a cut-off threshold.

Mentor: Prof. Joerg Sander

Software Energy Consumption Prediction

Developed machine learning models (i.e. regression models such as linear regression, svm regression, etc.) to predict energy consumption of software applications based on the system calls they make. The motivation of the work was to encourage developing energy efficient software applications.

Mentors: Prof. Russel Greiner / Prof. Abram Hindle

RNA Secondary Structure Prediction

Proposed a method to predict the secondary structures of RNA molecules using Conditional Random Fields (CRF). CRF models are mainly used in NLP and Biology to predict the labels of structured sequences. In contrast to ordinary classifiers advanced CRF models such as Skip-Chain CRF can take the dependency information with neighboring objects into account when predicting labels.

Mentors: Prof. Russel Greiner / Dr. Hosna Jabbari

KFGNN Queries in Time-Dependent Road Networks

Intorudced a novel alternate form of KNN queries, KFGNN (K-Fastest Group Nearest Neighbors), to query for k-set of moving objects in a road network which can reach a static query point fastest. Proposed algorithms to query KFGNN in static and time-dependent road networks. Proposed algorithms can gurantee optimal solution while achieving a 100x CPU permorance increase compared to a baseline approach.

Mentors: Prof. Mario Nascimento

Protein Interaction Network Analysis [Honors thesis]

Developed a community detection algorithm using PageRank network centrality measure to find key proteins in protein interaction networks and use them to find communities or clusters of proteins which show homogeneous behavriours. The proposed method outperforms others in recall rate while maintaining a competitive precision rate. [in media]

Advisor: Dr. Ruvan Weerasinghe

Virtual Eye

Project “Virtual Eye” is a platform which consists of a smart mobile client providing cost efficient and ubiquitous smart interaction with virtual environments deployed in a remote server. [in media]

Mentors: Dr. Chamath Keppitiyagama / Prof. Nihala Kodikara

Publications

My research works which have been published in peer-reviewed conferences and journals.