A Comparison of Classifiers for Intelligent Machine Usage Prediction

Probability estimation of machine usages is an essential task to the development of an intelligent device/environment. In this paper, we propose a generic framework to the task using the sliding window technique and incremental feature selection. The methodology is applied to a real-life dataset of office printers and the performances of different standard classifiers in supervised learning are compared. We conclude that Logistic Regression (LR) outperform other classifiers and is appropriate for the proposed framework. The use of Generic Bayesian Network (GBN) classifier is also promising, if combined with feature reduction methods.