Temporal data classification and rule extraction using a probabilistic decision tree

Abstract

Temporal data classification is an evolving area in machine learning and data mining in which time is included in learning procedure. In some real domains, observations are recorded on a time basis, so that there is a time sequence among the observation records. In this study, to make use of this temporal sequence, a procedure called temporalisation is applied to merge consecutive records. The learning algorithm is an entropy-based decision tree integrated with temporal decision tree concept. Furthermore, a probabilistic approach based on Bayes' theorem is applied to enhance prediction accuracy. The proposed temporal classifier is evaluated with three real datasets. It achieves better prediction results than an ordinary decision tree and produces temporal decision rules or temporal relationships.