An Evaluation of Discretization Techniques for HMM-Based Classifiers

Abstract

Discretization of continuous features is an important task to handle problems with real values in machine learning. Many supervised classification algorithms perform well using a discrete space and the discretization process of the continuous features is suitable for more traditional algorithms the process. In this paper, we present the classification model based on the Hidden Markov Model (HMM) developed recently by Benyacoub and al using several discretization methods existing in the literature to construct the classifier. We conduct an experimental study using 9 benchmarking data sets to evaluate the performance and examine the effect of discretization methods on the assessment of the proposed learning algorithm. {We conduct an experimental study using 9 benchmarking data sets to evaluate the performance and examine the effect of discretization methods on the assessment of the proposed learning algorithm. We report Accuracy (ACC) and Area Under the Curve (AUC), and we validate the global and pairwise differences across methods using the Friedman test followed by the Nemenyi post-hoc procedure.