THE DETERMINING OF OUTLIERS ON E-LEARNING DATA IN THE CONTEXT OF EDUCATIONAL DATA MINING AND LEARNING ANALYTICS

In the process of learning analytics, the determination of outliers and making smoothing before the analysis stage has an important place in reaching the right patterns. The outliers can be determined in the real-time, as well as, at the end of the data collection process. In this study, the use of outlier detection methods is discussed using educational data from an e-learning environment. Also, the methods were tested on a real-time system. The Moodle, Learning Management System (LMS) log records were used as the data set. The study group consists of 65 students. In this study, the total interaction times in hypertext, video, assessment, scorm, and forum themes were used as data set. Box-plot, Z, Grubbs, Rosner and Hampel methods were used to determine the outliers. Outliers are determined by processing through manual calculations without using the existing packaged software. At the same time, in order to evaluate integrability of these methods into the e-learning environment, some PHP script examples are coded by researchers. As a result of analyzes, it was shown that outlier numbers changed according to the methods. When the experiences obtained therefrom and database structure are considered; Z and Box-Plot methods are easier to implement in e-learning systems, for the real-time outlier detection than other methods. In other words, it has been seen that these methods are more functional in machine teaching. However, it should be noted that other methods have significant advantages, for that they require hypothesis test and give more sensitive results. In the context of machine learning, the positive and negative characteristics of these methods are discussed