Missing values (MVs) have been a persistent challenge in real-world datasets. In many prior studies, MVs have been removed without considering their underlying patterns, potentially discarding important information. This study has investigated seven machine learning-based MV handling methods and compared them with approaches reported in the literature, analyzing research published between 2019 and 2024. Using stratified ten-fold cross-validation, twelve classification algorithms, including standalone and ensemble models, have been evaluated on a binary classification problem with eleven performance metrics. The results have shown that combining missing value imputation (MVI) with data balancing (DB) substantially improved model performance. Accuracy increased from 0.9095 (without MVI and DB) to 0.9964 (with both), while the meta-learning metric Pmean rose from 0.6232 to 0.9954. These findings have demonstrated the critical role of effective data preparation, particularly MVI and DB, in enhancing predictive accuracy.