Predicting Lecturer’s Performance Using Datamining Techniques Based on Lecturer’s Characteristics and Historical Student Evaluation of Lecturer

Abstract

Abstract One of the most expensive resources in the higher educational process is lecturers, so most of the educational institutes spend a lot of effort and consume the HR resources to allocate the best lecturer to their students. Which will maximize the learning potential of the students according to the lecturer’s qualifications, skills and abilities. Therefore, the challenge is how to predict lecturer’s performance based on lecturer characterstics ak2nd historical student assessments of previous lecturers. Our approach uses data mining techniques to analyze existing data composed of lecturer academic and non-academic characteristics and predict prospective lecturer’s performance in order to support the decision-making in lecturer selection. We use four datamining techniques: Decision tree, K-Nearest Neighbor, Multinomial Logistic Regression and Naïve Bayesian. The models are trained and evaluated on a subset of the data. The model with the highest prediction outcome is selected. We used data belonging to the academic staff of Islamic University – Gaza (IUG) that taught in 11 semesters (from second semester 2011/2012 to summer semester 2014/2015). The dataset contains an attribute for the overall evaluation result from the end-of-semester questionnaires routinely filled by students, and 28 attributes of lecturer characteristics. The overall student questionnaire result is aggregated over all sections of a course that is taught by the lecturer in one semester. Based on training and evaluation of the four techniques mentioned above, and if we suppose that the closest prediction of the true evaluation is true, we can say that the models is predicting the evaluation truly or far from true in one step as next: Multinomial Logistic Regression gave the highest accuracy of 86.5%. Decision tree, K-Nearest Neighbor, and Naïve Bayesian gave accuracies of 85.0%, 86.4%, and 80.9%, respectively. Key words: Prediction, Lecturer Performance, Datamining, Lecturer’s Characteristics, Lecturers Evaluation