Rizk, NouhadZeng, Victor2019-01-022019-01-022018-10-18http://hdl.handle.net/10657/3782Early warning systems, or early alert systems, are systems to identify students at risk of failing a course. These systems use two categories of indicators: Traditional indicators such as assignment grades and class attendance, and “soft” factors such as the student’s behavior and learning network. Naturally, in the interest of preserving user engagement, an early warning system should ask the least amount of questions possible. In this research, we seek to determine if it is possible for an academic early warning system to obtain a level of prediction accuracy from an incomplete data set like that which can be obtained from a complete data set. A set of questions is developed about the student’s study habits, study attitudes, study anxiety, time management, learning network, and class participation. The questions answers are used to identify student’s characteristics. The pilot study is based on previous sessions of the Data Structures course at the University of Houston. First, classifiers are constructed based on two different algorithms k-nearest neighbors and feed-forward-neural network algorithms. Then training datasets of assignment and exams grades are measured using the three-fold cross validation method. In the future, we plan on implementing the study by asking the students in the upcoming fall session of Data Structures these questions and perform a mutual information analysis of their responses. If there is a high level of mutual information we will perform offline experiments on the data set to explore a mutual information approach and a PCA based approach to select optimal subsets of questions to ask individual students.en-USThe author of this work is the copyright owner. UH Libraries and the Texas Digital Library have their permission to store and provide access to this work. Further transmission, reproduction, or presentation of this work is prohibited except with permission of the author(s).Poster