Sparse, interpretable and transparent predictive model identification for healthcare data analysis

Data-driven modelling approaches play an indispensable role in analyzing and understanding complex processes. This study proposes a type of sparse, interpretable and transparent (SIT) machine learning model, which can be used to understand the dependent relationship of a response variable on a set of potential explanatory variables. An ideal candidate for such a SIT representation is the well-known NARMAX (nonlinear autoregressive moving average with exogenous inputs) model, which can be established from measured input and output data of the system of interest, and the final refined model is usually simple, parsimonious and easy to interpret. The performance of the proposed SIT models is evaluated through two real healthcare datasets.