Analisis Hasil Imputasi Menggunakan Arsitektur Imputasi Autoencoder

Abstract

Missing values in multivariate time series data are a critical issue in many domains, especially in healthcare datasets such as MIMIC-IV. This study aims to analyze the performance of imputation results using an Autoencoder-based architecture. Autoencoder is a deep learning model capable of learning data representations and reconstructing missing values through latent feature extraction. The research methodology includes data preprocessing, missing value simulation, model training, and evaluation using metrics such as MAE, RMSE, and R². The results show that Autoencoder-based imputation provides competitive performance in reconstructing missing values, particularly in nonlinear and complex patterns. However, the model's performance depends on the proportion of missing data and network architecture design. This study contributes to understanding the effectiveness of Autoencoder in multivariate time series imputation and provides a baseline for further development using hybrid models.