Identifying the Most Predictive Attributes Among Employability Signals of Undergraduate Students

Identifying an effective and scalable approach for student career management is one of the top priorities of Higher Education Institutions (HEI) through their career centers. Machine learning in the form of predictive analytics helps in achieving this goal. The ability to predict which students will most likely to be employed and why is imperative. This paper is focused on identifying why a student is most likely to be employed based on the career center data. In other words, the paper aims to determine in a scientific way which among the input attributes of are most predictive. The datasets used are from the career center of the Technological Institute of the Philippines and its school registrar. Datasets involved 27,000 information consists of (3,000 observations and 9 attributes) of students' mock job interview evaluation results, students' performance rating from the internship and general percentage average (GP A) of students enrolled from 2015 to 2018. To determine the most predictive attributes, three methods are used and compared. These are univariate selection (US), recursive feature elimination (RFE) and principal component analysis (PCA). Results show that mental alertness and manner of speaking are the two most predictive attributes. This study will help enhance the quality of career coaching by identifying students in need of improvement and tailoring guidance to help them.