Improving the expressiveness of black-box models for predicting student performance

Abstract

Early prediction systems of student performance can be very useful to guide student learning. For a prediction model to be really useful as an effective aid for learning, it must provide tools to adequately interpret progress, to detect trends and behaviour patterns and to identify the causes of learning problems. White-box and black-box techniques have been described in literature to implement prediction models. White-box techniques require a priori models to explore, which make them easy to interpret but difficult to be generalized and unable to detect unexpected relationships between data. Black-box techniques are easier to generalize and suitable to discover unsuspected relationships but they are cryptic and difficult to be interpreted for most teachers. In this paper a black-box technique is proposed to take advantage of the power and versatility of these methods, while making some decisions about the input data and design of the classifier that provide a rich output data set. A set of graphical tools is also proposed to exploit the output information and provide a meaningful guide to teachers and students. From our experience, a set of tips about how to design a prediction system and the representation of the output information is also provided.