Abstract

Predicting online instructors’ performance can help higher education institutions find issues and problems in the learning and teaching process as early as possible, giving them timely interventions to help ensure the quality of educational services. One of the significant challenges of evaluating and predicting online instructors’ performance is to determine the key indicators from a large amount of data generated from the learning management system (LMS). The recent advancements in the continuous development of machine learning technology have led to a new momentum of teaching performance prediction in online education. In this paper, we follow the design science research methodology. Firstly, a dataset is collected from a Midwest university LMS platform, and we use a wrapper-based method to explore a simple and effective prediction model to select four sets of key influence variables. Then, we fit, tune, and compare various machine learning prediction models on different selected variable sets. Finally, we suggest that the Random Forest (RF) on eight selected variables with the Synthetic Minority Over-sampling Technique (SMOTE) is the best prediction model based on our work.

Download

Share

COinS