Abstract

Traditional prediction methods are time-consuming and limited in identifying students at academic risk in a timely manner. On the other hand, Learning Analytics has certain advantages. The aim of this study was to analyze characteristics of predictive models based on learning analytics in higher education. A systematic review of Web of Science, Scopus and Eric databases was conducted using the keywords "learning analytics" and "prediction". Twelve research studies that met the inclusion criteria were selected. The results indicated that 100% of the studies sought to predict academic performance, including analytical, sociodemographic and sociocognitive variables as predictors. The most used learning management system was Moodle for blended learning and online courses. The studies were mainly developed in Europe; the samples consisted of up to 500 participants from Engineering and Technology. The most frequent type of analysis was regression in R and SPSS softwares. Most achieved a large prediction model (R2 > .30). It was concluded that the current construction of dropout prediction models has important limitations.