Missing Data Imputation Lab: Mean, KNN & Regression Fill

Real datasets arrive with holes, and how you fill them changes every model built downstream. This simulator generates a correlated three-feature synthetic dataset, hides a chosen fraction of one feature either completely at random (MCAR) or in a way that depends on another observed feature (MAR), and then reconstructs the missing values with mean, median, k-nearest-neighbour or linear-regression imputation — or removes the incomplete rows entirely. Because the dataset is synthetic, the true hidden values are known: each reconstructed point is drawn next to a translucent ghost of its real value, and live RMSE and variance-ratio readouts show exactly how much error and variance distortion each strategy introduces, especially once the missingness itself becomes biased.