A Quantitative Study of the Effect of Missing Data in Classifiers

In data mining approaches, predictive classification has a wide range of application. However, there are always missing data in the datasets, which affect the accuracy of classifiers. This paper investigates the influence of missing data to classifier. The sensitivity analysis of six classifiers to missing data is studied in experiments. The results indicate that, in the datasets, when the proportion of missing data exceeds 20%, they do have a huge adverse effect on the prediction accuracy. Among the six classifiers, the naive Bayesian classifier is the least sensitive to missing data. For the popular missing data treatment methods using prediction model to handle missing data, naive Bayesian classifier is preferred