Exploring Missing Value Handling Techniques for Optimized KNN Heart Disease Prediction

Heart disease prediction stands at the forefront of contemporary medical research, with machine learning models such as K-Nearest Neighbors (KNN) being increasingly employed for this purpose. A critical factor influencing the performance of these models is the presence of missing data, which, if not addressed appropriately, can severely hinder predictive accuracy. This study provides a comparative analysis of different missing value handling techniques and explores the impact of each method in predicting heart disease by applying the KNN model. Several imputation methods were employed in the existing literature and techniques such as mean imputation, median imputation, mode imputation, KNN imputation, list-wise deletion, and replacement with a constant value were scrutinized. The popular Framingham heart disease dataset is selected for the study and the missing values are addressed using the techniques mentioned. Subsequently, the KNN method is used to ascertain the predictive accuracy and this highlights the deviations in the results depending upon the chosen imputation method. Remarkably, the method of replacing missing values with a predefined constant value of zero achieved a more accurate result. The study focuses on the distinct characteristics, performance, suitable data type and limitations of each missing value handing technique and it provides valuable information for practitioners and researchers who aim to maximize the effectiveness of machine learning algorithms in cardiovascular disease prediction.