Abstract
The present paper presents a relatively new non-linear method to predict academic achievement of high school students, integrating the fields of psychometrics and machine learning. A sample composed by 135 high-school students (10th grade, 50.34% boys), aged between 14 and 19 years old (M = 15.44, DP = 1.09), answered to three psychological instruments: the Inductive Reasoning Developmental Test (TDRI), the Metacognitive Control Test (TCM) and the Brazilian Learning Approaches Scale (BLAS-Deep Approach). The first two tests have a self-appraisal scale attached, so we have five independent variables. The students’ responses to each test/scale were analyzed using the Rasch model. A subset of the original sample was created in order to separate the students in two balanced classes, high achievement (n = 41) and low achievement (n = 47), using grades from nine school subjects. In order to predict the class membership a machine learning non-linear model named Random Forest was used. The subset with the two classes was randomly split into two sets (training and testing) for cross validation. The result of the Random Forest showed a general accuracy of 75%, a specificity of 73.69% and a sensitivity of 68% in the training set. In the testing set, the general accuracy was 68.18%, with a specificity of 63.63% and with a sensitivity of 72.72%. The most important variable in the prediction was the TDRI. Finally, implications of the present study to the field of educational psychology were discussed.
Highlights
Machine learning is a relatively new science field composed by a broad class of computational and statistical methods to make predictions, inferences, and to discover new relations in data (Flach, 2012; Hastie, Tibshirani, & Friedman, 2009)
The present paper investigates the prediction of academic achievement of high-school students using two psychological tests and one educational scale: the Inductive Reasoning Developmental Test (TDRI), the Metacognitive Control Test (TCM) and the Brazilian Learning Approaches Scale (BLAS-Deep approach)
Considering the common language effect size, the probability that a the inductive reasoning developmental stage (TDRI) score taken at random from the high achievement group is greater than a TDRI score taken at random of the low achievement group is 73.41%
Summary
Machine learning is a relatively new science field composed by a broad class of computational and statistical methods to make predictions, inferences, and to discover new relations in data (Flach, 2012; Hastie, Tibshirani, & Friedman, 2009). There are several types of algorithms to perform classification and regression (Hastie et al, 2009) Among these algorithms, the tree based models are supervised learning techniques of special interest to the psychology and to the education research field. It can be used to discover which variable, or combination of variables, better predicts a given outcome, e.g. high or low academic achievement. It can identify the cutoff points for each variable that maximally predict the outcome, and can be applied to study the non-linear interaction effects of the independent variables and its relation to the quality of the prediction (Golino & Gomes, 2014). There are a growing number of applications of the tree-based models in different areas, from ADHA diagnosis (Eloyan et al, 2012; Skogli et al, 2013) to perceived stress (Scott, Jackson, & Bergeman, 2011), suicidal behavior (Baca-Garcia et al, 2007; Kuroki & Tilley, 2012), adaptive depression assessment (Gibbons et al, 2013), emotions (Tian et al, 2014; van der Wal & Kowalczyk, 2013) and education (Blanch & Aluja, 2013; Cortez & Silva, 2008; Golino & Gomes, 2014; Hardman, Paucar-Caceres, & Fielding, 2013)
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
