Decision Trees-based Anomaly Detection in Computer Assessment Results

Abstract

A survey of the current software in the area of computer assessments of students at a university is done. This work shows ways to improve quality of computer assessments and educational management. It suggests options to build a computer system for detecting anomalous assessment results of individual students and entire disciplines. This information could then be used to amend lectures and instructional material or management and better intercommunication to students. This work suggests feature sets for the analysis of student assessment results and demonstrates their availability in the current university software. After surveying known artificial intelligence systems this work proves the choice of decision trees for this problem. It suggests methods and algorithms to improve positive prediction rate of decision trees based on ideas of bootstrap method. A software tool was developed that implements the suggested algorithms.