Abstract
Dealing with missing values in data is an important feature engineering task in data science to prevent negative impacts on machine learning classification models in terms of accurate prediction. However, it is often unclear what the underlying cause of the missing values in real-life data is or rather the missing data mechanism that is causing the missingness. Thus, it becomes necessary to evaluate several missing data approaches for a given dataset. In this paper, we perform a comparative study of several approaches for handling missing values in data, namely listwise deletion, mean, mode, k–nearest neighbors, expectation-maximization, and multiple imputations by chained equations. The comparison is performed on two real-world datasets, using the following evaluation metrics: Accuracy, root mean squared error, receiver operating characteristics, and the F1 score. Most classifiers performed well across the missing data strategies. However, based on the result obtained, the support vector classifier method overall performed marginally better for the numerical data and naïve Bayes classifier for the categorical data when compared to the other evaluated missing value methods.