Breast cancer classification with missing data imputation

Abstract

Missing Data (MD) is a common drawback when applying Data Mining on breast cancer datasets since it affects the ability of the Data mining classifier. This study evaluates the influence of MD on three classifiers: Decision tree C4.5, Support vector machine (SVM), and Multi-Layer Perceptron (MLP). For this purpose, 162 experiments were conducted using KNN imputation with three missingness mechanisms (MCAR, MAR and NMAR), and nine percentages (form 10% to 90%) applied on two Wisconsin breast cancer datasets. The MD percentage affects negatively the classifier performance. MLP achieved the lowest accuracy rates regardless the MD mechanism/percentage.