Differential Entropy Estimation via One-Class SVM

This paper introduces the use of Support Vector Machine for entropy estimation of continuous random variables with well-defined probability density function. The method is based on support estimation and can converge to Shannon entropy or zero-order Rényi entropy depending of effective support set delimited. Simulated results indicate that the method proposed for effective support characterization gives asymptotically good results to Shannon entropy estimation in comparative with other three estimators based on: histogram, kernel smoothing and neighbor distances.