Algoritmo KNN basado en información mutua para clasificación de patrones con valores perdidos

Abstract

Incomplete data is a common drawback in reallife classification problems. Missing values in data sets may have different origins such as death of patients, equipment malfunctions, refusal of respondents to answer certain questions, and so on. This work1 presents an effective and robust approach for classification with unknown input data. In particular, an enhanced version of the K-Nearest Neighbours algorithm using Mutual Information is proposed. Results on two classification datasets shows the usefullness of this approach.