Missing Data Imputation using Multiple Imputation with Adaptive LASSO for Parkinson’s Disease Data

Abstract

Nerve cells are the brain's building blocks for the nervous system. Once destroyed, they do not regenerate. When these nerve cells are damaged, the dopamine they contain is depleted, causing motor abilities and speech to deteriorate. Before the brain cells are impaired, the voice goes through a series of modifications. Voice shifts assist in the early detection of Parkinson's disease, avoiding injury to brain cells that would result in decreased balance and movement. However, this condition often suffers from missed data in clinical outcomes due to a variety of factors such as dropout, illness, and so on. Hence, imputation of these kinds of missing data is always performed prior to performing an intent-to-treat study. Indeed, predictive analysis of data relating to disease will not be feasible without the use of a suitable framework that efficiently manages missing data. The paper proposes an Adaptive LASSO Imputation approach based on item answer theory, which allows multiple imputations to be done when working with multiple sources of correlation. The accuracy of each imputation procedure was assessed using the Root Mean Square Error (RMSE) and Mean Absolute Error. The proposed method is applied on three types of Missing Data i.e MAR, MCAR, NMAR. The outcomes demonstrated that the suggested approach outperforms all other algorithms.