Addressing Missing Data in Machine Learning: A Comparative Insights on Imputation Techniques and Classification Accuracy

Missing data presents a significant challenge in data analysis, affecting the quality of research outcomes. Real-world datasets often cover incomplete information due to various factors such as manual entry errors, equipment malfunctions, or inaccuracies in measurements, which complicates the data analysis process. Since most data analysis algorithms require complete datasets to function effectively, addressing missing data is crucial. Several strategies exist to handle missing data, yet selecting the best method can be challenging due to limited knowledge about each approach. This study categorizes machine learning methods for handling missing data into three key types: ignoring techniques, model-based techniques, and imputation techniques, detailing benefits and limitations of each method. The research systematically examines the influence of different accusation methods median imputation, forward fill, and backward fill on classification accuracy across several algorithms, including Random Forest Classifier, Decision Tree Classifier, MLP Classifier, Gaussian NB, SVC, and K Neighbors Classifier. Results indicate that median imputation generally offers high accuracy across most algorithms, while forward and backward fill techniques yield optimum results under specific conditions. The study underscores the importance of selecting the correct imputation approach to improve classification accuracy. Immediate tables are provided to compare the effectiveness of each imputation technique across different algorithms and datasets, offering valuable insights for choosing the most appropriate method.