Automated performance prediction for personalised learning

  • Samara Asam Banno

Student thesis: Doctoral Thesis


Designing an Intelligent Tutoring System (ITS) that simulates human learning with regard to different knowledge levels is a challenge. Most developed ITSs typically focus on either the building or the testing phase without paying appropriate consideration to the design phase. The result is that these systems offer specific choices in isolation, and can thus prove difficult to apply in situations where multiple factors interact. Researchers have therefore developed tutoring models that did include the design phase in their implementation. However these models do not consider the cognitive factors of students. Such ITS models lack the ability to provide accurate estimations as they do not analyse the students’ individual skills against the item skills, particularly when the learning items require multiple skills, which thus reduce student’s learning efficiency due to an incomplete representation of the student’s knowledge.
This thesis proposes a novel tutoring model, called ‘Cognitive Factor Analysis’ (CFA), which adapts student cognitive factors, such as guessing/slipping parameters and student proficiency levels, together with each item’s parameters to produce a better estimation of student latent performance.. CFA also introduces the concept of the Q-matrix from psychometrics and connects this to the students’ prior scores. The model does not only predict the students’ performance, but helps students to target the strengths and weaknesses in their knowledge levels. Therefore, CFA has an adverse impact on the student’s learning curve and reduce the student’s learning time by controlling the amount of time spent practicing the skill several times. It assumes the role of modelling the student’s learning by making inferences about their latent performance with multiple skills assessments.
CFA also contributes to cognitive learning psychology by exploring how computational models can be used to understand human behaviour. It shows how data generated from tutoring systems can be analysed and modelled to create and improve a unified computational theory of human learning. Furthermore, it encapsulates psychological findings in a format that can be used by instructional designers and educational scientists to support the development of tutoring systems.
Date of AwardDec 2018
Original languageEnglish
SupervisorLinda Yang (Supervisor)

Cite this