-
Prevalence and Correlation of Hypothyroidism With Pregnancy Outcomes Among Lebanese Women, Journal of the Endocrine Society, 2017
Assessment of hypothyroidism prevalence and clinical significance among pregnant women in Lebanon. We performed a single-center retrospective cohort study at the American University of Beirut Medical Center. Clinical, demographic, and laboratory data were collected and analyzed using trimester-specific ranges for hypothyroidism.
​
-
New Approach to Detect the Political Opinion in Tweets, Proceedings of the Joint Conference on Robotics and Intelligent Systems (JCRIS2014),Sao Paulo, Brazil, 2014
​
Detecting political opinion from tweets. We used tweets collected during the French presidential elections of 2012 and applied supervised machine learning techniques to help determine whether an author of a tweet is with or against a particular candidate.
​
-
Neighborhood Random Classification, Proceedings of the Geometric Science of Information 2013 conference (GSI 2013), Paris, France, 2013
​
We proposed an approach based on neighborhood graphs like relative neighborhood graphs (RNGs) or Gabriel graphs (GGs). Neighborhood graphs have never been introduced into EM approaches before. The results of our algorithm : Neighborhood Random Classification are very promising as they are equal to the best EM approaches such as Random Forest or those based on SVMs and many other tracks for improvement are still unexplored yet. In this preliminary and experimental work, we provide the methodological approach and many comparative results. We also provide some results on the influence of neighborhood structure regarding the efficiency of the classifier and draw some issues that deserves to be studied.
​
-
Topological learning with geometric graphs, Proceedings of the Joint Meeting of the IASC Conference, Yonsei University, Seoul, Korea, 2013
​​
Topological learning is a framework for machine learning where we care more about the topology of the data set than about the metric issues. Basically, the topology is expressed via the neighborhood structure that could be induced by the metric, defined using the representation space of observations, or could be given by the user, without any reference to a specific representation space. In this paper, we show how this framework might help to deal with some issues in machine learning.
​
-
Neighborhood Random Classification , Proceedings of the 16th Pacific-Asia Conference on Knowledge Discovery and Data Mining (PAKDD) 2012, Kualalampur, Malaysia, 2012
​​
Ensemble methods (EMs) have proved their efficiency in data mining, especially in supervised machine learning (ML). An EM generates a set of classifiers using one or several machine learning algorithms (MLAs) and aggregates them into a single classifier (Meta-Classifier, MC) using, for example, a majority rule vote. Instance Based (IB) MLAs , such as k-Nearest Neighbors (kNN), are very popular because of their straightforwardness. To implement these, it is simply necessary to define a dissimilarity measure based on the set of observations and fix the value of k. Thus, use of the kNN principle, as an EM algorithm is immediate. However, handling the parameter k might be difficult for some users. To simplify this problem, we can use approaches based on neighborhood graphs as alternatives. For example, relative neighborhood graphs (RNGs) or Gabriel graphs (GGs) are ”good” candidates. Like kNN, for an unlabeled observation, the classifier assigns a label based on neighborhood graphs according to the labels in the neighborhood. For example, we can simply use the majority rule vote in the neighborhood of the unlabeled observation. While many studies using kNN in the context of EMs have been done, we found no studies that assess the interestingness of neighborhood graphs in EM approaches. In this paper, we provide comparisons with many EM approaches based either on IB learning or on other methods such as kSVM, decision tree (random forest) etc.
​
-
Structural Equation Modeling: Theoretical Aspects and Applications using Data about the Social Status of Deaf Women in Lebanon, Master's Thesis, 2010
​
This thesis explored the concept of structural equation Modeling (SEM). Two estimation methods, Lisrel and PLS, are developed and compared. A real data, the status of Deaf Women in Lebanon 2005, is used to apply SEM using LISREL8.8 program. The result was a model that draws a relationship between the state of the deaf woman in Lebanon and her psychological state and personality.